โ†

๐Ÿ“„ Earnings Call Transcript ๋ฒˆ์—ญ ๊ฒฐ๊ณผ

?๋ฑค Presentation

Original Translation
MiniMax Group Inc. (MNMXF) Q4 2025 Earnings Call March 2, 2026 7:00 AM EST

Company Participants

Junjie Yan - Founder, Executive Chairman, CTO & CEO

Conference Call Participants

Alex Yao - JPMorgan Chase & Co, Research Division
Kenneth Fong - UBS Investment Bank, Research Division
Thomas Chong - Jefferies LLC, Research Division

Presentation

Operator

Good day, ladies and gentlemen. Thank you for standing by. Welcome to the MiniMax 2025 Full Year Financial Results Conference Call. Please note that English simultaneous interpretation will be provided for management's prepared remarks and Chinese Q&A session. [Operator Instructions]

I will now turn the call over to [ Ms.
**๋ฏธ๋‹ˆ๋งฅ์Šค ๊ทธ๋ฃน (์ฃผ) (MNMXF) 2025๋…„ 4๋ถ„๊ธฐ ์‹ค์  ๋ฐœํ‘œ ์ปจํผ๋Ÿฐ์Šค ์ฝœ**
2026๋…„ 3์›” 2์ผ, ๋™๋ถ€ ํ‘œ์ค€์‹œ ์˜ค์ „ 7์‹œ

**ํšŒ์‚ฌ ์ฐธ์„์ž**

์˜Œ์ฅ”์ œ - ์„ค๋ฆฝ์ž, ์ด์‚ฌํšŒ ์˜์žฅ, CTO ๊ฒธ CEO

**์ปจํผ๋Ÿฐ์Šค ์ฝœ ์ฐธ์„์ž**

์•Œ๋ ‰์Šค ์•ผ์˜ค - JP๋ชจ๊ฑด ์ฒด์ด์Šค, ๋ฆฌ์„œ์น˜ ๋ถ€๋ฌธ
์ผ€๋„ค์Šค ํ - UBS ํˆฌ์ž์€ํ–‰, ๋ฆฌ์„œ์น˜ ๋ถ€๋ฌธ
ํ† ๋งˆ์Šค ์ด - ์ œํ”„๋ฆฌ์Šค, ๋ฆฌ์„œ์น˜ ๋ถ€๋ฌธ

**๋ฐœํ‘œ**

**์‚ฌํšŒ์ž**
์‹ ์‚ฌ ์ˆ™๋…€ ์—ฌ๋Ÿฌ๋ถ„, ์•ˆ๋…•ํ•˜์‹ญ๋‹ˆ๊นŒ. ์ž ์‹œ ๊ธฐ๋‹ค๋ ค ์ฃผ์…”์„œ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๋ฏธ๋‹ˆ๋งฅ์Šค 2025๋…„ ์—ฐ๊ฐ„ ์žฌ๋ฌด ์‹ค์  ์ปจํผ๋Ÿฐ์Šค ์ฝœ์— ์˜ค์‹  ๊ฒƒ์„ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค. ๊ฒฝ์˜์ง„์˜ ์ค€๋น„๋œ ๋ฐœ์–ธ๊ณผ ์ค‘๊ตญ์–ด ์งˆ์˜์‘๋‹ต ์„ธ์…˜์—๋Š” ์˜์–ด ๋™์‹œ ํ†ต์—ญ์ด ์ œ๊ณต๋  ์˜ˆ์ •์ž„์„ ์•Œ๋ ค๋“œ๋ฆฝ๋‹ˆ๋‹ค. [์‚ฌํšŒ์ž ์ง€์นจ]

์ด์ œ [์„ฑํ•จ]๋‹˜๊ป˜ ๋งˆ์ดํฌ๋ฅผ ๋„˜๊ธฐ๊ฒ ์Šต๋‹ˆ๋‹ค.
Meredith Yu ], Director of IR at MiniMax. Unknown Executive

Thank you, operator. Good evening, and good morning to everyone. Welcome to MiniMax 2025 Full Year Financial Results Conference Call. Before we start, please note that today's discussion may contain forward-looking statements, which involve a number of risks and uncertainties. Actual results and outcomes may differ from those discussed. The company does not undertake any obligation to update any forward-looking information, except as required by law.
**๋ฉ”๋Ÿฌ๋””์Šค ์œ  (๋ฏธ๋‹ˆ๋งฅ์Šค IR ๋‹ด๋‹น ์ด์‚ฌ)**

**๋ฏธ์ƒ ์ž„์›**

์˜คํผ๋ ˆ์ดํ„ฐ๋‹˜, ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.

์ฐธ์„ํ•ด ์ฃผ์‹  ์—ฌ๋Ÿฌ๋ถ„, ์•ˆ๋…•ํ•˜์‹ญ๋‹ˆ๊นŒ? ์ข‹์€ ์ €๋…, ์ข‹์€ ์•„์นจ์ž…๋‹ˆ๋‹ค. ๋ฏธ๋‹ˆ๋งฅ์Šค์˜ 2025๋…„ ์—ฐ๊ฐ„ ์žฌ๋ฌด ์‹ค์  ์ปจํผ๋Ÿฐ์Šค ์ฝœ์— ์˜ค์‹  ๊ฒƒ์„ ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค.

๋ณธ ์ปจํผ๋Ÿฐ์Šค ์ฝœ์„ ์‹œ์ž‘ํ•˜๊ธฐ์— ์•ž์„œ, ๊ธˆ์ผ ๋…ผ์˜์—๋Š” ์—ฌ๋Ÿฌ ๊ฐ€์ง€ ์œ„ํ—˜๊ณผ ๋ถˆํ™•์‹ค์„ฑ์„ ์ˆ˜๋ฐ˜ํ•˜๋Š” ๋ฏธ๋ž˜ ์˜ˆ์ธก ์ง„์ˆ ์ด ํฌํ•จ๋  ์ˆ˜ ์žˆ์Œ์„ ์œ ์˜ํ•ด ์ฃผ์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค. ์‹ค์ œ ๊ฒฐ๊ณผ ๋ฐ ์„ฑ๊ณผ๋Š” ๋…ผ์˜๋œ ๋‚ด์šฉ๊ณผ ๋‹ค๋ฅผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋‹น์‚ฌ๋Š” ๋ฒ•๋ฅ ์— ์˜ํ•ด ์š”๊ตฌ๋˜๋Š” ๊ฒฝ์šฐ๋ฅผ ์ œ์™ธํ•˜๊ณ ๋Š” ์–ด๋– ํ•œ ๋ฏธ๋ž˜ ์˜ˆ์ธก ์ •๋ณด๋„ ์—…๋ฐ์ดํŠธํ•  ์˜๋ฌด๋ฅผ ์ง€์ง€ ์•Š์Šต๋‹ˆ๋‹ค.
For important information about this call, including forward-looking statements, please refer to the company's public information or 2025 full year results announcement ended December 31, 2025, issued earlier today. During today's call, management will also discuss certain non-IFRS financial measures. These are provided for additional information and should not replace IFRS-based financial results. For a definition of non-IFRS financial measures and reconciliation of IFRS to non-IFRS financial results and related risk factors, please refer to our 2025 full year results announcement. For today's call, management will use Chinese as the main language.๋ณธ ์ปจํผ๋Ÿฐ์Šค ์ฝœ๊ณผ ๊ด€๋ จ๋œ ์ค‘์š”ํ•œ ์ •๋ณด(๋ฏธ๋ž˜ ์˜ˆ์ธก ์ง„์ˆ  ํฌํ•จ)๋Š” ๋‹น์‚ฌ์˜ ๊ณต๊ฐœ ์ •๋ณด ๋˜๋Š” ๊ธˆ์ผ ์•ž์„œ ๋ฐœํ‘œ๋œ 2025๋…„ 12์›” 31์ผ ๋งˆ๊ฐ 2025 ํšŒ๊ณ„์—ฐ๋„ ์ „์ฒด ์‹ค์  ๋ฐœํ‘œ๋ฅผ ์ฐธ์กฐํ•˜์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค. ๊ธˆ์ผ ์ปจํผ๋Ÿฐ์Šค ์ฝœ์—์„œ ๊ฒฝ์˜์ง„์€ ํŠน์ • ๋น„-IFRS ์žฌ๋ฌด ์ง€ํ‘œ์— ๋Œ€ํ•ด์„œ๋„ ๋…ผ์˜ํ•  ์˜ˆ์ •์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์ถ”๊ฐ€ ์ •๋ณด ์ œ๊ณต ๋ชฉ์ ์œผ๋กœ ์ œ๊ณต๋˜๋ฉฐ, IFRS ๊ธฐ์ค€ ์žฌ๋ฌด ์‹ค์ ์„ ๋Œ€์ฒดํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ๋น„-IFRS ์žฌ๋ฌด ์ง€ํ‘œ์˜ ์ •์˜, IFRS์™€ ๋น„-IFRS ์žฌ๋ฌด ์‹ค์  ๊ฐ„์˜ ์กฐ์ • ๋‚ด์—ญ, ๊ทธ๋ฆฌ๊ณ  ๊ด€๋ จ ์œ„ํ—˜ ์š”์ธ์— ๋Œ€ํ•ด์„œ๋Š” ๋‹น์‚ฌ์˜ 2025 ํšŒ๊ณ„์—ฐ๋„ ์ „์ฒด ์‹ค์  ๋ฐœํ‘œ๋ฅผ ์ฐธ์กฐํ•˜์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค. ๋ณธ ์ปจํผ๋Ÿฐ์Šค ์ฝœ์—์„œ ๊ฒฝ์˜์ง„์€ ์ค‘๊ตญ์–ด๋ฅผ ์ฃผ์š” ์–ธ์–ด๋กœ ์‚ฌ์šฉํ•  ์˜ˆ์ •์ž…๋‹ˆ๋‹ค.
A third-party interpreter will provide simultaneous English interpretation in prepared remarks session and Q&A session. Please note that English interpretation is for convenience purposes only. In the case of any discrepancy, management's statements in the original language will prevail. Lastly, unless otherwise stated, all currency units mentioned are in USD. I will now hand the call over to our Founder and CEO of MiniMax, Dr. Yan Junjie. Junjie Yan
Founder, Executive Chairman, CTO & CEO

Dear investors and analysts, good evening. This is Yan Junjie. Thank you all for attending the first earnings call following our IPO.
์‚ฌ์ „ ๋ฐœ์–ธ ์„ธ์…˜๊ณผ ์งˆ์˜์‘๋‹ต ์„ธ์…˜์—์„œ๋Š” ์ œ3์ž ํ†ต์—ญ์‚ฌ๊ฐ€ ์˜์–ด ๋™์‹œ ํ†ต์—ญ์„ ์ œ๊ณตํ•  ์˜ˆ์ •์ž…๋‹ˆ๋‹ค. ์˜์–ด ํ†ต์—ญ์€ ํŽธ์˜๋ฅผ ์œ„ํ•œ ๊ฒƒ์ž„์„ ์œ ์˜ํ•˜์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค. ๋ถˆ์ผ์น˜ ๋ฐœ์ƒ ์‹œ์—๋Š” ๊ฒฝ์˜์ง„์˜ ์›์–ด ๋ฐœ์–ธ์ด ์šฐ์„ ํ•ฉ๋‹ˆ๋‹ค. ๋งˆ์ง€๋ง‰์œผ๋กœ, ๋ณ„๋„๋กœ ๋ช…์‹œ๋˜์ง€ ์•Š๋Š” ํ•œ, ์–ธ๊ธ‰๋˜๋Š” ๋ชจ๋“  ํ†ตํ™” ๋‹จ์œ„๋Š” ๋ฏธ๊ตญ ๋‹ฌ๋Ÿฌ(USD) ๊ธฐ์ค€์ž…๋‹ˆ๋‹ค.

์ด์ œ ๋ฏธ๋‹ˆ๋งฅ์Šค์˜ ์„ค๋ฆฝ์ž ๊ฒธ CEO์ด์‹  ์˜Œ์ฅ”์ง€์— ๋ฐ•์‚ฌ๋‹˜๊ป˜ ๋งˆ์ดํฌ๋ฅผ ๋„˜๊ธฐ๊ฒ ์Šต๋‹ˆ๋‹ค.

**์˜Œ์ฅ”์ง€์— (์„ค๋ฆฝ์ž, ํšŒ์žฅ, CTO ๊ฒธ CEO)**

ํˆฌ์ž์ž ๋ฐ ์• ๋„๋ฆฌ์ŠคํŠธ ์—ฌ๋Ÿฌ๋ถ„, ์•ˆ๋…•ํ•˜์‹ญ๋‹ˆ๊นŒ. ์˜Œ์ฅ”์ง€์—์ž…๋‹ˆ๋‹ค. ์ €ํฌ IPO ์ดํ›„ ์ฒซ ์‹ค์  ๋ฐœํ‘œ ์ปจํผ๋Ÿฐ์Šค์ฝœ์— ์ฐธ์„ํ•ด ์ฃผ์…”์„œ ์ง„์‹ฌ์œผ๋กœ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค.
I would like to take the opportunity of today's earnings call to share our progress over the past year and our strategic priorities for the next phase of growth. Let me start with a look back at 2025. For MiniMax, the key theme of this year was solidifying our foundation. In 2025, we built full-modality R&D capabilities, with global competitive models now in place across key modalities, including language, video, speech and music. Meanwhile, we continue to upgrade our products through ongoing technological innovation. This includes our enterprise and developer-facing Open Platform as well as customer products, such as MiniMax Agent, Hailuo AI, Talkie and Xingye.์˜ค๋Š˜ ์‹ค์  ๋ฐœํ‘œ ์ž๋ฆฌ๋ฅผ ๋นŒ๋ ค ์ง€๋‚œ ํ•œ ํ•ด ๋™์•ˆ์˜ ์ €ํฌ์˜ ์„ฑ๊ณผ์™€ ๋‹ค์Œ ์„ฑ์žฅ ๋‹จ๊ณ„๋ฅผ ์œ„ํ•œ ์ „๋žต์  ์šฐ์„ ์ˆœ์œ„๋ฅผ ๊ณต์œ ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค.

๋จผ์ € 2025๋…„์„ ๋˜๋Œ์•„๋ณด๋ฉฐ ๋ง์”€๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค. MiniMax์—๊ฒŒ 2025๋…„์€ ์ €ํฌ์˜ ๊ธฐ๋ฐ˜์„ ๊ณต๊ณ ํžˆ ํ•˜๋Š” ๊ฒƒ์ด ํ•ต์‹ฌ ์ฃผ์ œ์˜€์Šต๋‹ˆ๋‹ค. 2025๋…„์—๋Š” ์–ธ์–ด, ๋น„๋””์˜ค, ์Œ์„ฑ, ์Œ์•…์„ ํฌํ•จํ•œ ์ฃผ์š” ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ์ „๋ฐ˜์— ๊ฑธ์ณ ๊ธ€๋กœ๋ฒŒ ๊ฒฝ์Ÿ๋ ฅ์„ ๊ฐ–์ถ˜ ๋ชจ๋ธ๋“ค์„ ๊ตฌ์ถ•ํ•˜๋ฉฐ ํ’€ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(full-modality) R&D ์—ญ๋Ÿ‰์„ ํ™•๋ณดํ–ˆ์Šต๋‹ˆ๋‹ค.

ํ•œํŽธ, ์ €ํฌ๋Š” ์ง€์†์ ์ธ ๊ธฐ์ˆ  ํ˜์‹ ์„ ํ†ตํ•ด ์ œํ’ˆ์„ ๊พธ์ค€ํžˆ ์—…๊ทธ๋ ˆ์ด๋“œํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์—๋Š” ๊ธฐ์—… ๋ฐ ๊ฐœ๋ฐœ์ž ๋Œ€์ƒ์˜ ์˜คํ”ˆ ํ”Œ๋žซํผ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ MiniMax Agent, Hailuo AI, Talkie, Xingye์™€ ๊ฐ™์€ ๊ณ ๊ฐ์šฉ ์ œํ’ˆ๋“ค์ด ํฌํ•จ๋ฉ๋‹ˆ๋‹ค.
We also made further progress in deepening our global footprint. In large language models, during the fourth quarter of last year, we launched 3 updated models, M2, M2.1 and M2-her. M2 redefined the balance among performance, cost and speed and incorporated 3 key capabilities: coding, tool use and deep search. Its performance has approached the leading global standards. Following its release, M2 saw rapid adoption within the global developer community, becoming the first Chinese model on OpenRouter to exceed 50 billion tokens in daily consumption while ranking first on the HuggingFace global trending leaderboard during the week.์ €ํฌ๋Š” ๋˜ํ•œ ๊ธ€๋กœ๋ฒŒ ์ž…์ง€ ๊ฐ•ํ™”์—๋„ ์ถ”๊ฐ€์ ์ธ ์ง„์ „์„ ์ด๋ฃจ์—ˆ์Šต๋‹ˆ๋‹ค. ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ(LLM) ๋ถ„์•ผ์—์„œ๋Š” ์ž‘๋…„ 4๋ถ„๊ธฐ์— M2, M2.1, M2-her ๋“ฑ 3์ข…์˜ ์—…๋ฐ์ดํŠธ ๋ชจ๋ธ์„ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. M2๋Š” ์„ฑ๋Šฅ, ๋น„์šฉ, ์†๋„ ๊ฐ„์˜ ๊ท ํ˜•์„ ์žฌ์ •์˜ํ–ˆ์œผ๋ฉฐ, ์ฝ”๋”ฉ, ๋„๊ตฌ ์‚ฌ์šฉ, ๋”ฅ ์„œ์น˜ ๋“ฑ 3๊ฐ€์ง€ ํ•ต์‹ฌ ๊ธฐ๋Šฅ์„ ํƒ‘์žฌํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ ์„ฑ๋Šฅ์€ ์„ธ๊ณ„ ์ตœ๊ณ  ์ˆ˜์ค€์— ๊ทผ์ ‘ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ถœ์‹œ ์ดํ›„ M2๋Š” ๊ธ€๋กœ๋ฒŒ ๊ฐœ๋ฐœ์ž ์ปค๋ฎค๋‹ˆํ‹ฐ ๋‚ด์—์„œ ๋น ๋ฅด๊ฒŒ ์ฑ„ํƒ๋˜์—ˆ์œผ๋ฉฐ, OpenRouter์—์„œ ์ผ์ผ ํ† ํฐ ์†Œ๋น„๋Ÿ‰ 500์–ต ๊ฐœ๋ฅผ ๋ŒํŒŒํ•œ ์ตœ์ดˆ์˜ ์ค‘๊ตญ ๋ชจ๋ธ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋™์‹œ์— ํ•ด๋‹น ์ฃผ๊ฐ„ HuggingFace ๊ธ€๋กœ๋ฒŒ ํŠธ๋ Œ๋”ฉ ๋ฆฌ๋”๋ณด๋“œ์—์„œ 1์œ„๋ฅผ ๊ธฐ๋กํ–ˆ์Šต๋‹ˆ๋‹ค.
Building on M2, we quickly launched M2.1 with a focus on improving performance on complex, real-world tasks, particularly in coding and workplace scenarios, where it demonstrated stronger capabilities in understanding and executing multistep instructions. Additionally, M2-her serves as the foundation model supporting our AI interactive products, namely Xingye and Talkie. It is designed to deliver more natural and personalized conversational experiences and was ranked first globally in overall performance in 100-turn long-context dialogue testing.M2๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ, ๋‹น์‚ฌ๋Š” ๋ณต์žกํ•œ ์‹ค์ œ ํ™˜๊ฒฝ ์ž‘์—…, ํŠนํžˆ ์ฝ”๋”ฉ ๋ฐ ์—…๋ฌด ํ™˜๊ฒฝ ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ์˜ ์„ฑ๋Šฅ ๊ฐœ์„ ์— ์ค‘์ ์„ ๋‘๊ณ  M2.1์„ ์‹ ์†ํ•˜๊ฒŒ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. M2.1์€ ๋‹ค๋‹จ๊ณ„ ์ง€์‹œ๋ฅผ ์ดํ•ดํ•˜๊ณ  ์‹คํ–‰ํ•˜๋Š” ๋ฐ ์žˆ์–ด ๋”์šฑ ๊ฐ•๋ ฅํ•œ ์—ญ๋Ÿ‰์„ ์ž…์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, M2-her๋Š” ๋‹น์‚ฌ์˜ AI ๋Œ€ํ™”ํ˜• ์ œํ’ˆ์ธ Xingye์™€ Talkie๋ฅผ ์ง€์›ํ•˜๋Š” ๊ธฐ๋ฐ˜ ๋ชจ๋ธ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋”์šฑ ์ž์—ฐ์Šค๋Ÿฝ๊ณ  ๊ฐœ์ธํ™”๋œ ๋Œ€ํ™” ๊ฒฝํ—˜์„ ์ œ๊ณตํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์œผ๋ฉฐ, 100ํ„ด ์žฅ๋ฌธ ๋งฅ๋ฝ ๋Œ€ํ™” ํ…Œ์ŠคํŠธ์—์„œ ์ „๋ฐ˜์ ์ธ ์„ฑ๋Šฅ ๋ถ€๋ฌธ์—์„œ ์ „ ์„ธ๊ณ„ 1์œ„๋ฅผ ์ฐจ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค.
In February, we released M2.5, which achieved global leading performance across key productivity scenarios, including coding, tool use and workplace applications. In coding, M2.5 set a new industry record on the SWE-bench Verified benchmark while delivering a 37% efficiency improvement compared with the previous generation, M2.1. More importantly, M2.5 makes the operation of complex agents economically viable. Running continuously for 1 hour at an out -- speed of 100 tokens per second costs only USD 1. This means that a budget of $10,000, 4 agents can operate continuously for an entire year.2์›”์— ์ €ํฌ๋Š” M2.5๋ฅผ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. M2.5๋Š” ์ฝ”๋”ฉ, ๋„๊ตฌ ์‚ฌ์šฉ, ์—…๋ฌด์šฉ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ํฌํ•จํ•œ ์ฃผ์š” ์ƒ์‚ฐ์„ฑ ์‹œ๋‚˜๋ฆฌ์˜ค ์ „๋ฐ˜์—์„œ ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ์„ ๋„์ ์ธ ์„ฑ๋Šฅ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ M2.5๋Š” SWE-bench Verified ๋ฒค์น˜๋งˆํฌ์—์„œ ์ƒˆ๋กœ์šด ์—…๊ณ„ ๊ธฐ๋ก์„ ์„ธ์› ์œผ๋ฉฐ, ์ด์ „ ์„ธ๋Œ€์ธ M2.1๊ณผ ๋น„๊ตํ•˜์—ฌ 37%์˜ ํšจ์œจ์„ฑ ํ–ฅ์ƒ์„ ๋‹ฌ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค.

๋” ์ค‘์š”ํ•˜๊ฒŒ๋Š”, M2.5๋Š” ๋ณต์žกํ•œ ์—์ด์ „ํŠธ์˜ ์šด์˜์„ ๊ฒฝ์ œ์ ์œผ๋กœ ์‹คํ–‰ ๊ฐ€๋Šฅํ•˜๊ฒŒ ๋งŒ๋“ญ๋‹ˆ๋‹ค. ์ดˆ๋‹น 100ํ† ํฐ์˜ ์†๋„๋กœ 1์‹œ๊ฐ„ ๋™์•ˆ ์—ฐ์†์ ์œผ๋กœ ์‹คํ–‰ํ•˜๋Š” ๋ฐ ๋‹จ 1๋‹ฌ๋Ÿฌ๋งŒ ์†Œ์š”๋ฉ๋‹ˆ๋‹ค. ์ด๋Š” 1๋งŒ ๋‹ฌ๋Ÿฌ์˜ ์˜ˆ์‚ฐ์œผ๋กœ 4๊ฐœ์˜ ์—์ด์ „ํŠธ๊ฐ€ 1๋…„ ๋‚ด๋‚ด ์—ฐ์†์ ์œผ๋กœ ์ž‘๋™ํ•  ์ˆ˜ ์žˆ์Œ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.
Breakthroughs in model capability have also driven rapid growth in usage with M2.5, quickly topping the OpenRouter rankings following its release. From M2 to M2.1, now M2.5, each generation has delivered significant improvements in both capability and adoption. In February 2026, average daily token consumption across the M2 text model series was more than 6x the level recorded in December 2025, with the token consumption from Coding Plan growing by more than tenfold. On the multimodal front, we have now established a model coverage across video, speech and music.๋ชจ๋ธ ์—ญ๋Ÿ‰์˜ ํš๊ธฐ์ ์ธ ๋ฐœ์ „์€ M2.5์˜ ์‚ฌ์šฉ๋Ÿ‰ ๊ธ‰์ฆ์„ ๊ฒฌ์ธํ–ˆ์œผ๋ฉฐ, M2.5๋Š” ์ถœ์‹œ ์งํ›„ OpenRouter ์ˆœ์œ„์—์„œ ๋น ๋ฅด๊ฒŒ 1์œ„๋ฅผ ์ฐจ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค. M2๋ถ€ํ„ฐ M2.1, ๊ทธ๋ฆฌ๊ณ  ํ˜„์žฌ M2.5์— ์ด๋ฅด๊ธฐ๊นŒ์ง€, ๊ฐ ์„ธ๋Œ€๋Š” ์—ญ๋Ÿ‰๊ณผ ์ฑ„ํƒ๋ฅ  ๋ชจ๋‘์—์„œ ์ƒ๋‹นํ•œ ๊ฐœ์„ ์„ ์ด๋ฃจ์–ด๋ƒˆ์Šต๋‹ˆ๋‹ค. 2026๋…„ 2์›”, M2 ํ…์ŠคํŠธ ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ์˜ ์ผํ‰๊ท  ํ† ํฐ ์†Œ๋น„๋Ÿ‰์€ 2025๋…„ 12์›” ๊ธฐ๋ก๋œ ์ˆ˜์ค€๋ณด๋‹ค 6๋ฐฐ ์ด์ƒ ์ฆ๊ฐ€ํ–ˆ์œผ๋ฉฐ, ํŠนํžˆ ์ฝ”๋”ฉ ํ”Œ๋žœ(Coding Plan)์˜ ํ† ํฐ ์†Œ๋น„๋Ÿ‰์€ 10๋ฐฐ ์ด์ƒ ๊ธ‰์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ(multimodal) ๋ถ„์•ผ์—์„œ๋Š” ๋น„๋””์˜ค, ์Œ์„ฑ, ์Œ์•… ์ „๋ฐ˜์— ๊ฑธ์ณ ๋ชจ๋ธ ์ปค๋ฒ„๋ฆฌ์ง€๋ฅผ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค.
In October last year, we released our video model Hailuo 2.3, which delivered significant improvements in character motion, visual quality and stylistic expression. We also introduced a faster Fast model, which can reduce batch content creation costs by up to 50%. We further upgraded Media Agent within Hailuo AI, which supports full-modality content creation to generate the final output in one click. As of the end of 2025, our video models have helped creators worldwide generate more than 600 million videos in total. In October last year, we released our speech model, Speech 2.6, which was optimized for voice agent scenarios and significantly enhanced the voice interaction performance.์ž‘๋…„ 10์›”, ์ €ํฌ๋Š” ์บ๋ฆญํ„ฐ ๋™์ž‘, ์‹œ๊ฐ์  ํ’ˆ์งˆ ๋ฐ ์Šคํƒ€์ผ ํ‘œํ˜„์—์„œ ์ƒ๋‹นํ•œ ๊ฐœ์„ ์„ ์ด๋ฃฌ ๋น„๋””์˜ค ๋ชจ๋ธ 'ํ•ด์ผ๋กœ 2.3'์„ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ผ๊ด„ ์ฝ˜ํ…์ธ  ์ œ์ž‘ ๋น„์šฉ์„ ์ตœ๋Œ€ 50%๊นŒ์ง€ ์ ˆ๊ฐํ•  ์ˆ˜ ์žˆ๋Š” ๋” ๋น ๋ฅธ 'ํŒจ์ŠคํŠธ ๋ชจ๋ธ'๋„ ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค. 'ํ•ด์ผ๋กœ AI' ๋‚ด์˜ ๋ฏธ๋””์–ด ์—์ด์ „ํŠธ๋„ ํ•œ์ธต ๋” ์—…๊ทธ๋ ˆ์ด๋“œํ•˜์—ฌ, ํ’€ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(full-modality) ์ฝ˜ํ…์ธ  ์ƒ์„ฑ์„ ์ง€์›ํ•˜๊ณ  ํด๋ฆญ ํ•œ ๋ฒˆ์œผ๋กœ ์ตœ์ข… ๊ฒฐ๊ณผ๋ฌผ์„ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ํ–ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ ๋ง ๊ธฐ์ค€์œผ๋กœ, ์ €ํฌ ๋น„๋””์˜ค ๋ชจ๋ธ๋“ค์€ ์ „ ์„ธ๊ณ„ ํฌ๋ฆฌ์—์ดํ„ฐ๋“ค์ด ์ด 6์–ต ๊ฐœ ์ด์ƒ์˜ ๋น„๋””์˜ค๋ฅผ ์ƒ์„ฑํ•˜๋„๋ก ์ง€์›ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ž‘๋…„ 10์›”์—๋Š” ์Œ์„ฑ ์—์ด์ „ํŠธ ์‹œ๋‚˜๋ฆฌ์˜ค์— ์ตœ์ ํ™”๋˜์–ด ์Œ์„ฑ ์ƒํ˜ธ์ž‘์šฉ ์„ฑ๋Šฅ์„ ๋Œ€ํญ ๊ฐ•ํ™”ํ•œ ์Œ์„ฑ ๋ชจ๋ธ '์Šคํ”ผ์น˜ 2.6'์„ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค.
It achieved a global leading ultra-low latency and supports more than 40 languages. As of the end of last year, our speech model had helped users worldwide to generate over 200 million hours of speech in total, making it one of the core infrastructure platforms in the voice intelligence ecosystem. Our newly released music models, Music 2.0 and 2.5, also achieved significant advancements. They can reliably handle a wide range of vocal styles and emotional expressions. In the process of developing these models and products, we've also continuously advanced our AI-native organizational evolution.๊ธ€๋กœ๋ฒŒ ์„ ๋„์ ์ธ ์ดˆ์ €์ง€์—ฐ(ultra-low latency)์„ ๋‹ฌ์„ฑํ–ˆ์œผ๋ฉฐ, 40๊ฐœ ์ด์ƒ์˜ ์–ธ์–ด๋ฅผ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค. ์ž‘๋…„ ๋ง ๊ธฐ์ค€์œผ๋กœ, ๋‹น์‚ฌ์˜ ์Œ์„ฑ ๋ชจ๋ธ์€ ์ „ ์„ธ๊ณ„ ์‚ฌ์šฉ์ž๋“ค์ด ์ด 2์–ต ์‹œ๊ฐ„ ์ด์ƒ์˜ ์Œ์„ฑ์„ ์ƒ์„ฑํ•˜๋„๋ก ์ง€์›ํ•˜๋ฉฐ, ์Œ์„ฑ ์ธํ…”๋ฆฌ์ „์Šค ์ƒํƒœ๊ณ„์˜ ํ•ต์‹ฌ ์ธํ”„๋ผ ํ”Œ๋žซํผ ์ค‘ ํ•˜๋‚˜๋กœ ์ž๋ฆฌ๋งค๊น€ํ–ˆ์Šต๋‹ˆ๋‹ค.

์ƒˆ๋กญ๊ฒŒ ๊ณต๊ฐœ๋œ ๋‹น์‚ฌ์˜ ์Œ์•… ๋ชจ๋ธ์ธ Music 2.0 ๋ฐ 2.5 ๋˜ํ•œ ๊ด„๋ชฉํ•  ๋งŒํ•œ ๋ฐœ์ „์„ ์ด๋ฃจ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๋ชจ๋ธ๋“ค์€ ๊ด‘๋ฒ”์œ„ํ•œ ๋ณด์ปฌ ์Šคํƒ€์ผ๊ณผ ๊ฐ์ • ํ‘œํ˜„์„ ์•ˆ์ •์ ์œผ๋กœ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ชจ๋ธ๊ณผ ์ œํ’ˆ์„ ๊ฐœ๋ฐœํ•˜๋Š” ๊ณผ์ •์—์„œ, ์ €ํฌ๋Š” ๋˜ํ•œ AI ๋„ค์ดํ‹ฐ๋ธŒ(AI-native) ์กฐ์ง ์ง„ํ™”๋ฅผ ์ง€์†์ ์œผ๋กœ ์ถ”์ง„ํ•ด ์™”์Šต๋‹ˆ๋‹ค.
Internally, our agent interns now support nearly 90% of employees, with use cases spanning software development, data analysis, operations management, talent recruitment and sales and marketing. We view ourselves as a testing ground for the evolution of AI-native organizational capabilities, one that will steadily improve our R&D efficiency. In January this year, we productized these capabilities we had accumulated and released MiniMax Agent 2.0, enabling agents to directly access the users' local workspaces. At the same time, we will launch the Expert Agent feature, allowing users to create a domain-specific agents tailored to professional use cases.๋‚ด๋ถ€์ ์œผ๋กœ, ์ €ํฌ ์—์ด์ „ํŠธ ์ธํ„ด๋“ค์€ ํ˜„์žฌ ์ง์›๋“ค์˜ ๊ฑฐ์˜ 90%๋ฅผ ์ง€์›ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ํ™œ์šฉ ์‚ฌ๋ก€๋Š” ์†Œํ”„ํŠธ์›จ์–ด ๊ฐœ๋ฐœ, ๋ฐ์ดํ„ฐ ๋ถ„์„, ์šด์˜ ๊ด€๋ฆฌ, ์ธ์žฌ ์ฑ„์šฉ, ์˜์—… ๋ฐ ๋งˆ์ผ€ํŒ… ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์— ๊ฑธ์ณ ์žˆ์Šต๋‹ˆ๋‹ค. ์šฐ๋ฆฌ๋Š” ์Šค์Šค๋กœ๋ฅผ AI ๋„ค์ดํ‹ฐ๋ธŒ ์กฐ์ง ์—ญ๋Ÿ‰ ์ง„ํ™”๋ฅผ ์œ„ํ•œ ํ…Œ์ŠคํŠธ ๋ฒ ๋“œ๋กœ ๊ฐ„์ฃผํ•˜๋ฉฐ, ์ด๋Š” ์šฐ๋ฆฌ์˜ R&D ํšจ์œจ์„ฑ์„ ๊พธ์ค€ํžˆ ํ–ฅ์ƒ์‹œํ‚ฌ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์˜ฌํ•ด 1์›”, ์šฐ๋ฆฌ๋Š” ์ถ•์ ํ•ด ์˜จ ์ด๋Ÿฌํ•œ ์—ญ๋Ÿ‰๋“ค์„ ์ œํ’ˆํ™”ํ•˜์—ฌ MiniMax Agent 2.0์„ ์ถœ์‹œํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ์—์ด์ „ํŠธ๊ฐ€ ์‚ฌ์šฉ์ž์˜ ๋กœ์ปฌ ์ž‘์—… ๊ณต๊ฐ„์— ์ง์ ‘ ์ ‘๊ทผํ•  ์ˆ˜ ์žˆ๋„๋ก ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋™์‹œ์—, ์šฐ๋ฆฌ๋Š” Expert Agent ๊ธฐ๋Šฅ์„ ์ถœ์‹œํ•˜์—ฌ ์‚ฌ์šฉ์ž๋“ค์ด ์ „๋ฌธ์ ์ธ ํ™œ์šฉ ์‚ฌ๋ก€์— ๋งž์ถฐ ํŠน์ • ๋„๋ฉ”์ธ์— ํŠนํ™”๋œ ์—์ด์ „ํŠธ๋ฅผ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
As of the end of February, professional users had cumulatively created over 50,000 Expert Agents, addressing specialized challenges through deep knowledge and capability integration. Even speaking of the OpenClaw, which is very popular. Actually, even before the OpenClaw project gained broad attention, its founder, Peter, had already spoke highly of MiniMax models, describing the M2.1 model as his preferred and the best open source model. Following OpenClaw's official launch, the combined performance and cost advantages of the M2 series enabled more developers to adopt the models at significantly lower cost.2์›” ๋ง ๊ธฐ์ค€์œผ๋กœ ์ „๋ฌธ ์‚ฌ์šฉ์ž๋“ค์€ ๊นŠ์ด ์žˆ๋Š” ์ง€์‹๊ณผ ์—ญ๋Ÿ‰ ํ†ตํ•ฉ์„ ํ†ตํ•ด ์ „๋ฌธ์ ์ธ ๊ณผ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” 5๋งŒ ๊ฐœ ์ด์ƒ์˜ Expert Agent๋ฅผ ๋ˆ„์  ์ƒ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค. ๋งค์šฐ ์ธ๊ธฐ ์žˆ๋Š” OpenClaw์˜ ์‚ฌ๋ก€๋งŒ ๋ณด๋”๋ผ๋„ ๊ทธ๋ ‡์Šต๋‹ˆ๋‹ค. ์‚ฌ์‹ค, OpenClaw ํ”„๋กœ์ ํŠธ๊ฐ€ ํญ๋„“์€ ์ฃผ๋ชฉ์„ ๋ฐ›๊ธฐ ์ „๋ถ€ํ„ฐ ๊ทธ ์„ค๋ฆฝ์ž์ธ ํ”ผํ„ฐ๋Š” ์ด๋ฏธ MiniMax ๋ชจ๋ธ๋“ค์„ ๋†’์ด ํ‰๊ฐ€ํ–ˆ์œผ๋ฉฐ, M2.1 ๋ชจ๋ธ์„ ์ž์‹ ์ด ์„ ํ˜ธํ•˜๋Š” ์ตœ๊ณ ์˜ ์˜คํ”ˆ ์†Œ์Šค ๋ชจ๋ธ์ด๋ผ๊ณ  ๊ทน์ฐฌํ–ˆ์Šต๋‹ˆ๋‹ค. OpenClaw์˜ ๊ณต์‹ ์ถœ์‹œ ์ดํ›„, M2 ์‹œ๋ฆฌ์ฆˆ์˜ ๊ฒฐํ•ฉ๋œ ์„ฑ๋Šฅ ๋ฐ ๋น„์šฉ ์ด์  ๋•๋ถ„์— ๋” ๋งŽ์€ ๊ฐœ๋ฐœ์ž๋“ค์ด ํ›จ์”ฌ ๋” ๋‚ฎ์€ ๋น„์šฉ์œผ๋กœ ํ•ด๋‹น ๋ชจ๋ธ๋“ค์„ ์ฑ„ํƒํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
Our agent product also proactively supported OpenClaw, launching MaxClaw, which further lowered the barrier to entry for users. Next up, we would like to talk about our progress on monetization. Alongside our -- for the whole year, we generated USD 79 million in revenue for the full year, representing 159% year-over-year. Along this, revenue from AI-native products reached USD 53 million, up 143% year-over-year, while revenue from our Open Platform was around USD 26 million, up 198% year-over-year. We are seeing the revenue accelerating in 2025. For example, our Open Platform serving enterprise customers and individual developers.์ €ํฌ ์—์ด์ „ํŠธ ์ œํ’ˆ์€ OpenClaw๋ฅผ ์„ ์ œ์ ์œผ๋กœ ์ง€์›ํ•˜๋ฉฐ MaxClaw๋ฅผ ์ถœ์‹œํ•˜์—ฌ ์‚ฌ์šฉ์ž ์ง„์ž… ์žฅ๋ฒฝ์„ ๋”์šฑ ๋‚ฎ์ท„์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, ์ˆ˜์ตํ™”(monetization) ์ง„ํ–‰ ์ƒํ™ฉ์— ๋Œ€ํ•ด ๋ง์”€๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ์—ฐ๊ฐ„ ์ด 7,900๋งŒ ๋‹ฌ๋Ÿฌ์˜ ๋งค์ถœ์„ ๊ธฐ๋กํ–ˆ์œผ๋ฉฐ, ์ด๋Š” ์ „๋…„ ๋Œ€๋น„ 159% ์ฆ๊ฐ€ํ•œ ์ˆ˜์น˜์ž…๋‹ˆ๋‹ค. ์ด์™€ ํ•จ๊ป˜ AI ๋„ค์ดํ‹ฐ๋ธŒ ์ œํ’ˆ ๋งค์ถœ์€ 5,300๋งŒ ๋‹ฌ๋Ÿฌ์— ๋‹ฌํ•˜๋ฉฐ ์ „๋…„ ๋Œ€๋น„ 143% ์ฆ๊ฐ€ํ–ˆ๊ณ , ์˜คํ”ˆ ํ”Œ๋žซํผ ๋งค์ถœ์€ ์•ฝ 2,600๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ๊ธฐ๋กํ•˜๋ฉฐ ์ „๋…„ ๋Œ€๋น„ 198% ์ฆ๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ๋Š” 2025๋…„์— ๋งค์ถœ์ด ๋”์šฑ ๊ฐ€์†ํ™”๋  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ €ํฌ ์˜คํ”ˆ ํ”Œ๋žซํผ์€ ๊ธฐ์—… ๊ณ ๊ฐ๊ณผ ๊ฐœ์ธ ๊ฐœ๋ฐœ์ž๋“ค์„ ๋Œ€์ƒ์œผ๋กœ ์„œ๋น„์Šค๋ฅผ ์ œ๊ณตํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
Our new user registration in February 2026 were more than 4x the level recorded in December 2025. As of December 31, 2025, we had cumulatively served more than 236 million users across over 200 countries and regions as well as 214,000 enterprise customers and developers from more than 100 countries and regions. Revenue from international markets accounted for more than 70% of our total revenue in 2025. And international revenue represented over 50% of total revenue for our Open Platform. Since the release of M2.5, we have seen strong traction in international markets, attracting significant inbound interest from new global customers with positive word of mouth continue to build momentum.2026๋…„ 2์›” ์‹ ๊ทœ ์‚ฌ์šฉ์ž ๋“ฑ๋ก ๊ฑด์ˆ˜๋Š” 2025๋…„ 12์›”์— ๊ธฐ๋ก๋œ ์ˆ˜์ค€๋ณด๋‹ค 4๋ฐฐ ์ด์ƒ ์ฆ๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ 12์›” 31์ผ ๊ธฐ์ค€์œผ๋กœ, ๋‹น์‚ฌ๋Š” 200๊ฐœ ์ด์ƒ์˜ ๊ตญ๊ฐ€ ๋ฐ ์ง€์—ญ์—์„œ 2์–ต 3์ฒœ 6๋ฐฑ๋งŒ ๋ช… ์ด์ƒ์˜ ์‚ฌ์šฉ์ž์—๊ฒŒ ๋ˆ„์  ์„œ๋น„์Šค๋ฅผ ์ œ๊ณตํ–ˆ์œผ๋ฉฐ, 100๊ฐœ ์ด์ƒ์˜ ๊ตญ๊ฐ€ ๋ฐ ์ง€์—ญ์œผ๋กœ๋ถ€ํ„ฐ 21๋งŒ 4์ฒœ ๊ฐœ์˜ ๊ธฐ์—… ๊ณ ๊ฐ ๋ฐ ๊ฐœ๋ฐœ์ž๋ฅผ ํ™•๋ณดํ–ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ ํ•ด์™ธ ์‹œ์žฅ ๋งค์ถœ์€ ๋‹น์‚ฌ ์ด ๋งค์ถœ์˜ 70% ์ด์ƒ์„ ์ฐจ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ๋‹น์‚ฌ์˜ ์˜คํ”ˆ ํ”Œ๋žซํผ ์ด ๋งค์ถœ์—์„œ ํ•ด์™ธ ๋งค์ถœ์ด 50% ์ด์ƒ์„ ์ฐจ์ง€ํ–ˆ์Šต๋‹ˆ๋‹ค. M2.5 ์ถœ์‹œ ์ดํ›„, ๋‹น์‚ฌ๋Š” ํ•ด์™ธ ์‹œ์žฅ์—์„œ ๊ฐ•๋ ฅํ•œ ๊ฒฌ์ธ๋ ฅ์„ ๋ณด์˜€์œผ๋ฉฐ, ๊ธ์ •์ ์ธ ์ž…์†Œ๋ฌธ์ด ์ง€์†์ ์œผ๋กœ ํ™•์‚ฐ๋˜๋ฉด์„œ ์ƒˆ๋กœ์šด ๊ธ€๋กœ๋ฒŒ ๊ณ ๊ฐ๋“ค์˜ ์ƒ๋‹นํ•œ ๊ด€์‹ฌ์„ ์œ ์น˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
Leading global cloud providers and AI-native cloud platforms, including Google Vertex AI, Microsoft Azure, AI Foundry, Fireworks AI, and Nebius AI have deployed MiniMax models. We have also become the default model on leading coding platforms such as OpenCode and KiloCode. And early this morning, Notion launched M2.5, which is the first -- its first and only open source model option. In addition, while offering our services above, we further enhanced compute efficiency by driving engineering optimizations and deliver meaningful gains, benefiting from iterative improvements in algorithm optimization, operator implementation and encoding and decoding engineering.์„ ๋„์ ์ธ ๊ธ€๋กœ๋ฒŒ ํด๋ผ์šฐ๋“œ ์ œ๊ณต์—…์ฒด ๋ฐ AI ๋„ค์ดํ‹ฐ๋ธŒ ํด๋ผ์šฐ๋“œ ํ”Œ๋žซํผ(Google Vertex AI, Microsoft Azure, AI Foundry, Fireworks AI, Nebius AI ๋“ฑ)์—์„œ MiniMax ๋ชจ๋ธ์„ ๋ฐฐํฌํ–ˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ OpenCode ๋ฐ KiloCode์™€ ๊ฐ™์€ ์„ ๋„์ ์ธ ์ฝ”๋”ฉ ํ”Œ๋žซํผ์˜ ๊ธฐ๋ณธ ๋ชจ๋ธ๋กœ ์ฑ„ํƒ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์˜ค๋Š˜ ์•„์นจ ์ผ์ฐ, Notion์€ M2.5๋ฅผ ์ถœ์‹œํ–ˆ๋Š”๋ฐ, ์ด๋Š” Notion์˜ ์ตœ์ดˆ์ด์ž ์œ ์ผํ•œ ์˜คํ”ˆ ์†Œ์Šค ๋ชจ๋ธ ์˜ต์…˜์ž…๋‹ˆ๋‹ค.

์ด์™€ ๋”๋ถˆ์–ด, ์œ„์—์„œ ์–ธ๊ธ‰ํ•œ ์„œ๋น„์Šค๋“ค์„ ์ œ๊ณตํ•˜๋Š” ๋™์‹œ์—, ์ €ํฌ๋Š” ์—”์ง€๋‹ˆ์–ด๋ง ์ตœ์ ํ™”๋ฅผ ์ถ”์ง„ํ•˜์—ฌ ์ปดํ“จํŒ… ํšจ์œจ์„ฑ์„ ๋”์šฑ ํ–ฅ์ƒ์‹œ์ผฐ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ตœ์ ํ™”, ์—ฐ์‚ฐ์ž ๊ตฌํ˜„, ์ธ์ฝ”๋”ฉ ๋ฐ ๋””์ฝ”๋”ฉ ์—”์ง€๋‹ˆ์–ด๋ง ๋ถ„์•ผ์—์„œ์˜ ๋ฐ˜๋ณต์ ์ธ ๊ฐœ์„ ์„ ํ†ตํ•ด ์ƒ๋‹นํ•œ ์„ฑ๊ณผ๋ฅผ ๊ฐ€์ ธ์™”์Šต๋‹ˆ๋‹ค.
As of February 2026, inference computer cost per million tokens for the M2 text model series has declined by over 50% compared with the December 2025 levels. Over the same period, inference latency for Hailuo video generation model has decreased by more than 30%. As our model capabilities continue to iterate and improve, the benefits of scale have emerged. For the full year 2025, gross profit reached USD 20 million, up 437% year-over-year, with gross margin improving to 25.4%, up 13 percentage points from 12.2% in 2024.2026๋…„ 2์›” ๊ธฐ์ค€, M2 ํ…์ŠคํŠธ ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ์˜ 100๋งŒ ํ† ํฐ๋‹น ์ถ”๋ก  ์ปดํ“จํŒ… ๋น„์šฉ์€ 2025๋…„ 12์›” ์ˆ˜์ค€ ๋Œ€๋น„ 50% ์ด์ƒ ๊ฐ์†Œํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฐ™์€ ๊ธฐ๊ฐ„ ๋™์•ˆ, ํ•˜์ด๋ฃจ์˜ค(Hailuo) ์˜์ƒ ์ƒ์„ฑ ๋ชจ๋ธ์˜ ์ถ”๋ก  ์ง€์—ฐ ์‹œ๊ฐ„์€ 30% ์ด์ƒ ๊ฐ์†Œํ–ˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ ๋ชจ๋ธ ์—ญ๋Ÿ‰์ด ์ง€์†์ ์œผ๋กœ ๋ฐ˜๋ณต ๊ฐœ์„ ๋˜๋ฉด์„œ, ๊ทœ๋ชจ์˜ ์ด์ ์ด ๋‚˜ํƒ€๋‚˜๊ธฐ ์‹œ์ž‘ํ–ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ ํ•œ ํ•ด ๋™์•ˆ, ๋งค์ถœ์ด์ด์ต์€ 2์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ๊ธฐ๋กํ•˜๋ฉฐ ์ „๋…„ ๋Œ€๋น„ 437% ์ฆ๊ฐ€ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋งค์ถœ์ด์ด์ต๋ฅ ์€ 25.4%๋กœ ๊ฐœ์„ ๋˜์–ด 2024๋…„ 12.2%์—์„œ 13%ํฌ์ธํŠธ ์ƒ์Šนํ–ˆ์Šต๋‹ˆ๋‹ค.
On the expenses side, sales and marketing expenses decreased by 40% year-over-year, while R&D expenses increased by 33.8% year-over-year, though significantly below our revenue growth rate. For the full year 2025, adjusted net loss was USD 250 million. As commercialization continued to advance and model optimization drove cost efficiencies, our adjusted net loss margin narrowed significantly. In the first 2 months of 2026, we have already seen strong growth momentum. As of February 2026, our ARR has exceeded USD 150 million. Next up, I would like to share our outlook for the future. We believe that in 2026, intelligence levels will advance significantly.๋น„์šฉ ์ธก๋ฉด์—์„œ, ํŒ๋งค ๋ฐ ๋งˆ์ผ€ํŒ… ๋น„์šฉ์€ ์ „๋…„ ๋Œ€๋น„ 40% ๊ฐ์†Œํ–ˆ์œผ๋ฉฐ, R&D ๋น„์šฉ์€ ์ „๋…„ ๋Œ€๋น„ 33.8% ์ฆ๊ฐ€ํ–ˆ์ง€๋งŒ, ์ด๋Š” ์ €ํฌ ๋งค์ถœ ์„ฑ์žฅ๋ฅ ์— ๋น„ํ•ด ํ˜„์ €ํžˆ ๋‚ฎ์€ ์ˆ˜์ค€์ด์—ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ ์—ฐ๊ฐ„ ๊ธฐ์ค€์œผ๋กœ ์กฐ์ • ์ˆœ์†์‹ค์€ 2์–ต 5์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ๊ธฐ๋กํ–ˆ์Šต๋‹ˆ๋‹ค. ์ƒ์—…ํ™”๊ฐ€ ๊ณ„์† ์ง„์ „๋˜๊ณ  ๋ชจ๋ธ ์ตœ์ ํ™”๋ฅผ ํ†ตํ•ด ๋น„์šฉ ํšจ์œจ์„ฑ์ด ํ–ฅ์ƒ๋˜๋ฉด์„œ, ์ €ํฌ์˜ ์กฐ์ • ์ˆœ์†์‹ค๋ฅ ์€ ํฌ๊ฒŒ ์ถ•์†Œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. 2026๋…„ ์ฒซ ๋‘ ๋‹ฌ ๋™์•ˆ, ์ €ํฌ๋Š” ์ด๋ฏธ ๊ฐ•๋ ฅํ•œ ์„ฑ์žฅ ๋ชจ๋ฉ˜ํ…€์„ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค. 2026๋…„ 2์›” ๊ธฐ์ค€์œผ๋กœ, ์ €ํฌ์˜ ARR(์—ฐ๊ฐ„ ๋ฐ˜๋ณต ๋งค์ถœ)์€ 1์–ต 5์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ์ดˆ๊ณผํ–ˆ์Šต๋‹ˆ๋‹ค. ๋‹ค์Œ์œผ๋กœ, ์ €ํฌ์˜ ํ–ฅํ›„ ์ „๋ง์„ ๊ณต์œ ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” 2026๋…„์—๋Š” ์ง€๋Šฅ ์ˆ˜์ค€์ด ํฌ๊ฒŒ ๋ฐœ์ „ํ•  ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค.
Our own efforts will focus on the following 3 aspects: First, in software development, we expect to see the emergence of L4 to L5 levels of intelligence, marking the shift from AI as a tool to AI as a colleague-level collaborator. Second, across professional workplaces, we expect to see a pace of progress similar to what we saw in coding last year. In particular, the delivery capabilities and penetration of AI agents in workplace areas will improve meaningfully. Third, multimodal creation this year will move towards the direct generation of production-ready mid- to long-form content, with formats emerging that are increasingly closer to streaming and real-time output.์ €ํฌ๋Š” ๋‹ค์Œ ์„ธ ๊ฐ€์ง€ ์ธก๋ฉด์— ๋…ธ๋ ฅ์„ ์ง‘์ค‘ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์ฒซ์งธ, ์†Œํ”„ํŠธ์›จ์–ด ๊ฐœ๋ฐœ ๋ถ„์•ผ์—์„œ๋Š” L4์—์„œ L5 ์ˆ˜์ค€์˜ ์ง€๋Šฅ ์ถœํ˜„์„ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” AI๊ฐ€ ๋‹จ์ˆœํ•œ ๋„๊ตฌ๋ฅผ ๋„˜์–ด ๋™๋ฃŒ ์ˆ˜์ค€์˜ ํ˜‘๋ ฅ์ž๋กœ ์ „ํ™˜๋จ์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค.

๋‘˜์งธ, ์ „๋ฌธ ์—…๋ฌด ํ™˜๊ฒฝ ์ „๋ฐ˜์—์„œ๋Š” ์ž‘๋…„ ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ ๋ชฉ๊ฒฉํ–ˆ๋˜ ๊ฒƒ๊ณผ ์œ ์‚ฌํ•œ ๋ฐœ์ „ ์†๋„๋ฅผ ๊ธฐ๋Œ€ํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ, ์—…๋ฌด ์˜์—ญ์—์„œ AI ์—์ด์ „ํŠธ์˜ ์ œ๊ณต ์—ญ๋Ÿ‰๊ณผ ์นจํˆฌ์œจ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์…‹์งธ, ์˜ฌํ•ด ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ฐฝ์ž‘์€ ์ฆ‰์‹œ ํ™œ์šฉ ๊ฐ€๋Šฅํ•œ ์ค‘์žฅํŽธ ์ฝ˜ํ…์ธ ์˜ ์ง์ ‘์ ์ธ ์ƒ์„ฑ์„ ํ–ฅํ•ด ๋‚˜์•„๊ฐˆ ๊ฒƒ์ด๋ฉฐ, ์ŠคํŠธ๋ฆฌ๋ฐ ๋ฐ ์‹ค์‹œ๊ฐ„ ์ถœ๋ ฅ์— ์ ์  ๋” ๊ฐ€๊นŒ์›Œ์ง€๋Š” ํ˜•์‹๋“ค์ด ์ถœํ˜„ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.
Taken together, these 3 developments signal new technical challenges, a significant expansion in the supply of intelligence at scale and a substantial window of innovation at the application layer. They're also implying a meaningful increase in the demand placed on our platform, with the token volume likely to grow by 1 to 2 orders of magnitude. Our next-generation M3 and Hailuo 3 model series are designed with these needs in mind. In parallel, we are rapidly strengthening our infrastructure and continuing to attract top talent, shifting our focus from optimizing training efficiency alone to driving higher R&D and iteration efficiency.์ข…ํ•ฉ์ ์œผ๋กœ ๋ณผ ๋•Œ, ์ด ์„ธ ๊ฐ€์ง€ ๋ฐœ์ „์€ ์ƒˆ๋กœ์šด ๊ธฐ์ˆ ์  ๊ณผ์ œ, ๋Œ€๊ทœ๋ชจ ์ง€๋Šฅ ๊ณต๊ธ‰์˜ ์ƒ๋‹นํ•œ ํ™•์žฅ, ๊ทธ๋ฆฌ๊ณ  ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ณ„์ธต์—์„œ์˜ ์ƒ๋‹นํ•œ ํ˜์‹  ๊ธฐํšŒ๋ฅผ ์‹œ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด๋Š” ์šฐ๋ฆฌ ํ”Œ๋žซํผ์— ๋Œ€ํ•œ ์ˆ˜์š”์˜ ์˜๋ฏธ ์žˆ๋Š” ์ฆ๊ฐ€๋ฅผ ์˜๋ฏธํ•˜๋ฉฐ, ํ† ํฐ ๋ณผ๋ฅจ์€ 1~2 ์ž๋ฆฟ์ˆ˜ ๊ทœ๋ชจ๋กœ ์„ฑ์žฅํ•  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒ๋ฉ๋‹ˆ๋‹ค. ๋‹น์‚ฌ์˜ ์ฐจ์„ธ๋Œ€ M3 ๋ฐ Hailuo 3 ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ๋Š” ์ด๋Ÿฌํ•œ ํ•„์š”๋ฅผ ์—ผ๋‘์— ๋‘๊ณ  ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด์™€ ๋ณ‘ํ–‰ํ•˜์—ฌ, ๋‹น์‚ฌ๋Š” ์ธํ”„๋ผ๋ฅผ ์‹ ์†ํ•˜๊ฒŒ ๊ฐ•ํ™”ํ•˜๊ณ  ์ตœ๊ณ  ์ธ์žฌ๋ฅผ ๊ณ„์†ํ•ด์„œ ์œ ์น˜ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ํ›ˆ๋ จ ํšจ์œจ์„ฑ ์ตœ์ ํ™”์—๋งŒ ์ง‘์ค‘ํ•˜๋Š” ๊ฒƒ์—์„œ ๋ฒ—์–ด๋‚˜ ๋” ๋†’์€ R&D ๋ฐ ๋ฐ˜๋ณต ํšจ์œจ์„ฑ์„ ์ถ”์ง„ํ•˜๋Š” ๋ฐฉํ–ฅ์œผ๋กœ ์ดˆ์ ์„ ์ „ํ™˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
At the strategic level, we are evolving from a large model company into a platform company for the AI era. In the Internet era, platform companies primarily served as gateway for traffic. In the AI era, however, platform companies are those that define and advance new intelligence paradigms that are able to capture the products and commercial value created by those paradigm shifts. This requires stability to shape emerging intelligence frameworks to sustain innovation in both technology and products and to provide a scalable infrastructure and highly efficient token throughput capability.์ „๋žต์  ์ฐจ์›์—์„œ, ์ €ํฌ๋Š” ๋Œ€๊ทœ๋ชจ ๋ชจ๋ธ ๊ธฐ์—…์—์„œ AI ์‹œ๋Œ€๋ฅผ ์œ„ํ•œ ํ”Œ๋žซํผ ๊ธฐ์—…์œผ๋กœ ์ง„ํ™”ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ธํ„ฐ๋„ท ์‹œ๋Œ€์—๋Š” ํ”Œ๋žซํผ ๊ธฐ์—…๋“ค์ด ์ฃผ๋กœ ํŠธ๋ž˜ํ”ฝ์˜ ๊ด€๋ฌธ ์—ญํ• ์„ ํ–ˆ์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ AI ์‹œ๋Œ€์—๋Š” ํ”Œ๋žซํผ ๊ธฐ์—…์ด ์ƒˆ๋กœ์šด ์ง€๋Šฅ ํŒจ๋Ÿฌ๋‹ค์ž„์„ ์ •์˜ํ•˜๊ณ  ๋ฐœ์ „์‹œํ‚ค๋ฉฐ, ๊ทธ๋Ÿฌํ•œ ํŒจ๋Ÿฌ๋‹ค์ž„ ์ „ํ™˜์œผ๋กœ ์ธํ•ด ์ฐฝ์ถœ๋˜๋Š” ์ œํ’ˆ๊ณผ ์ƒ์—…์  ๊ฐ€์น˜๋ฅผ ํฌ์ฐฉํ•˜๊ณ  ํ™•๋ณดํ•  ์ˆ˜ ์žˆ๋Š” ๊ธฐ์—…์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด์„œ๋Š” ์ƒˆ๋กœ์šด ์ง€๋Šฅ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ํ˜•์„ฑํ•  ์ˆ˜ ์žˆ๋Š” ์•ˆ์ •์„ฑ, ๊ธฐ์ˆ  ๋ฐ ์ œํ’ˆ ํ˜์‹ ์„ ์ง€์†ํ•  ์ˆ˜ ์žˆ๋Š” ์—ญ๋Ÿ‰, ๊ทธ๋ฆฌ๊ณ  ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ์ธํ”„๋ผ์™€ ๊ณ ํšจ์œจ ํ† ํฐ ์ฒ˜๋ฆฌ๋Ÿ‰ ์—ญ๋Ÿ‰์„ ๊ฐ–์ถ”๋Š” ๊ฒƒ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
We believe that we are one of the few companies that we -- that have established and continue to strengthen these capabilities. So the value of an AI era platform company can be simply framed as the density of intelligence provided multiplied by token throughput. When both dimensions are sufficiently strong, platform value naturally emerges. Getting at this historical inflection point of industry, our confidence grounded in 2 factors. The acceleration of the AI industry is increasingly evident. Breakthrough in model capability, deployment of agent applications, and maturation monetization models are obviously continuing to expand at the industry selling.์ €ํฌ๋Š” ์ด๋Ÿฌํ•œ ์—ญ๋Ÿ‰์„ ๊ตฌ์ถ•ํ•˜๊ณ  ์ง€์†์ ์œผ๋กœ ๊ฐ•ํ™”ํ•ด ์˜จ ์†Œ์ˆ˜์˜ ๊ธฐ์—… ์ค‘ ํ•˜๋‚˜๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ AI ์‹œ๋Œ€ ํ”Œ๋žซํผ ๊ธฐ์—…์˜ ๊ฐ€์น˜๋Š” ์ œ๊ณต๋˜๋Š” ์ง€๋Šฅ์˜ ๋ฐ€๋„์— ํ† ํฐ ์ฒ˜๋ฆฌ๋Ÿ‰์„ ๊ณฑํ•œ ๊ฐ’์œผ๋กœ ๊ฐ„๋‹จํžˆ ์ •์˜๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ๋‘ ๊ฐ€์ง€ ์ฐจ์›์ด ์ถฉ๋ถ„ํžˆ ๊ฐ•๋ ฅํ•  ๋•Œ, ํ”Œ๋žซํผ ๊ฐ€์น˜๋Š” ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ์ฐฝ์ถœ๋ฉ๋‹ˆ๋‹ค.

์ด ์‚ฐ์—…์˜ ์—ญ์‚ฌ์ ์ธ ๋ณ€๊ณก์ ์— ์„œ์„œ, ์ €ํฌ์˜ ์ž์‹ ๊ฐ์€ ๋‘ ๊ฐ€์ง€ ์š”์ธ์— ๊ธฐ๋ฐ˜์„ ๋‘๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. AI ์‚ฐ์—…์˜ ๊ฐ€์†ํ™”๊ฐ€ ๋”์šฑ ๋ถ„๋ช…ํ•ด์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ ์—ญ๋Ÿ‰์˜ ํš๊ธฐ์ ์ธ ๋ฐœ์ „, ์—์ด์ „ํŠธ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์˜ ๋ฐฐํฌ, ๊ทธ๋ฆฌ๊ณ  ์ˆ˜์ตํ™” ๋ชจ๋ธ์˜ ์„ฑ์ˆ™์€ ๋ถ„๋ช…ํžˆ ์‚ฐ์—… ์ „๋ฐ˜์— ๊ฑธ์ณ ๊ณ„์†ํ•ด์„œ ํ™•์žฅ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
So we are already strong, seeing strong growth momentum. We're confident in becoming a core builder of the AI platform ecosystem. That concludes our prepared remarks. We're now ready to take your questions.๋”ฐ๋ผ์„œ ์ €ํฌ๋Š” ์ด๋ฏธ ๊ฐ•๋ ฅํ•œ ์ž…์ง€๋ฅผ ๋‹ค์ง€๊ณ  ์žˆ์œผ๋ฉฐ, ๊ฐ•๋ ฅํ•œ ์„ฑ์žฅ ๋ชจ๋ฉ˜ํ…€์„ ์ด์–ด๊ฐ€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. AI ํ”Œ๋žซํผ ์ƒํƒœ๊ณ„์˜ ํ•ต์‹ฌ ๊ตฌ์ถ•์ž๋กœ์„œ ์ž๋ฆฌ๋งค๊น€ํ•  ๊ฒƒ์ž„์„ ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค.

์ด์ƒ์œผ๋กœ ์ €ํฌ์˜ ๋ชจ๋‘ ๋ฐœ์–ธ์„ ๋งˆ์น˜๊ฒ ์Šต๋‹ˆ๋‹ค. ์ด์ œ ์งˆ์˜์‘๋‹ต ์‹œ๊ฐ„์„ ๊ฐ–๊ฒ ์Šต๋‹ˆ๋‹ค.

๐Ÿ“Œ ์š”์•ฝ

MiniMax Group Inc.์˜ 2025๋…„ 4๋ถ„๊ธฐ ์‹ค์  ๋ฐœํ‘œ ์š”์•ฝ (์ „๋ฌธ ํˆฌ์ž์ž์šฉ):

* **๊ฒฌ์กฐํ•œ ๋งค์ถœ ์„ฑ์žฅ ๋ฐ ํšจ์œจ์„ฑ ๊ฐœ์„  ์† ์ˆœ์†์‹ค ์ง€์†:** 2025๋…„ ์—ฐ๊ฐ„ ๋งค์ถœ์€ ์ „๋…„ ๋Œ€๋น„ 159% ์ฆ๊ฐ€ํ•œ 7,900๋งŒ ๋‹ฌ๋Ÿฌ, ์ด์ด์ต์€ 437% ์ฆ๊ฐ€ํ•œ 2,000๋งŒ ๋‹ฌ๋Ÿฌ(์ด๋งˆ์ง„ 25.4%)๋ฅผ ๊ธฐ๋กํ–ˆ์œผ๋‚˜, ์กฐ์ • ์ˆœ์†์‹ค์€ 2์–ต 5์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ๊ธฐ๋กํ–ˆ์Šต๋‹ˆ๋‹ค. 2026๋…„ 2์›” ๊ธฐ์ค€ ์—ฐ๊ฐ„๋ฐ˜๋ณต๋งค์ถœ(ARR)์€ 1์–ต 5์ฒœ๋งŒ ๋‹ฌ๋Ÿฌ๋ฅผ ์ดˆ๊ณผํ–ˆ์œผ๋ฉฐ, M2 ํ…์ŠคํŠธ ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ์˜ ์ถ”๋ก  ์ปดํ“จํŒ… ๋น„์šฉ์ด 50% ์ด์ƒ ์ ˆ๊ฐ๋˜๋Š” ๋“ฑ ํšจ์œจ์„ฑ ๊ฐœ์„  ๋…ธ๋ ฅ์ด ๋‹๋ณด์ž…๋‹ˆ๋‹ค.
* **๋ชจ๋ธ ์„ฑ๋Šฅ ๊ณ ๋„ํ™” ๋ฐ ๊ด‘๋ฒ”์œ„ํ•œ ์‹œ์žฅ ์ฑ„ํƒ:** M2.5 ๋ชจ๋ธ์€ ์ฝ”๋”ฉ ๋ฐ ์—…๋ฌด ์‹œ๋‚˜๋ฆฌ์˜ค์—์„œ ๊ธ€๋กœ๋ฒŒ ์„ ๋„์  ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ•˜๋ฉฐ ๋ณต์žกํ•œ AI ์—์ด์ „ํŠธ ์šด์˜์˜ ๊ฒฝ์ œ์„ฑ์„ ํ™•๋ณดํ–ˆ์Šต๋‹ˆ๋‹ค. 2025๋…„ 12์›” ๋Œ€๋น„ 2026๋…„ 2์›” M2 ํ…์ŠคํŠธ ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ์˜ ์ผ์ผ ํ† ํฐ ์†Œ๋น„๋Ÿ‰์€ 6๋ฐฐ ์ด์ƒ, ์ฝ”๋”ฉ ํ”Œ๋žœ์€ 10๋ฐฐ ์ด์ƒ ์ฆ๊ฐ€ํ–ˆ์œผ๋ฉฐ, Google, Microsoft, Notion ๋“ฑ ์ฃผ์š” ๊ธ€๋กœ๋ฒŒ ํ”Œ๋žซํผ์— ๋ชจ๋ธ์ด ๋ฐฐํฌ๋˜๋ฉฐ ์‹œ์žฅ ์ฑ„ํƒ์ด ๊ฐ€์†ํ™”๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
* **"AI ์‹œ๋Œ€์˜ ํ”Œ๋žซํผ ๊ธฐ์—…"์œผ๋กœ์˜ ์ „๋žต์  ์ „ํ™˜ ๋ฐ ๋ฏธ๋ž˜ ์„ฑ์žฅ ์ž์‹ ๊ฐ:** ํšŒ์‚ฌ๋Š” "AI ์‹œ๋Œ€์˜ ํ”Œ๋žซํผ ๊ธฐ์—…"์œผ๋กœ์˜ ์ „ํ™˜์„ ์„ ์–ธํ•˜๋ฉฐ, 2026๋…„ ์†Œํ”„ํŠธ์›จ์–ด ๊ฐœ๋ฐœ ๋ฐ ์ „๋ฌธ ์—…๋ฌด ๋ถ„์•ผ์—์„œ AI ์—์ด์ „ํŠธ์˜ ์ง€๋Šฅ ์ˆ˜์ค€ ๋ฐ ์นจํˆฌ์œจ ๊ธ‰์ฆ, ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ฝ˜ํ…์ธ  ์ƒ์„ฑ ์—ญ๋Ÿ‰ ๊ฐ•ํ™”๋ฅผ ์ „๋งํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด์— ๋”ฐ๋ผ ํ”Œ๋žซํผ ํ† ํฐ ๋ณผ๋ฅจ์ด 1~2 ์ž๋ฆฟ์ˆ˜ ๊ทœ๋ชจ๋กœ ์ฆ๊ฐ€ํ•  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒํ•˜๋ฉฐ, ์ฐจ์„ธ๋Œ€ M3 ๋ฐ Hailuo 3 ๋ชจ๋ธ ๊ฐœ๋ฐœ๊ณผ ์ธํ”„๋ผ ๊ฐ•ํ™”์— ์ง‘์ค‘ํ•  ๊ณ„ํš์ž…๋‹ˆ๋‹ค. ๊ฒฝ์˜์ง„์€ AI ํ”Œ๋žซํผ ์ƒํƒœ๊ณ„์˜ ํ•ต์‹ฌ ๋นŒ๋”๊ฐ€ ๋  ๊ฒƒ์ด๋ผ๋Š” ๊ฐ•ํ•œ ์ž์‹ ๊ฐ์„ ํ‘œ๋ช…ํ–ˆ์Šต๋‹ˆ๋‹ค.


??Q&A

Original Translation
Operator: [Operator Instructions] Please dial in Chinese line if you want to ask question. Your first question comes from [ Gary Yi ] of Morgan Stanley.**Operator:** ์งˆ๋ฌธํ•˜์‹ค ๋ถ„๊ป˜์„œ๋Š” ์ค‘๊ตญ์–ด ํšŒ์„ ์œผ๋กœ ์ ‘์†ํ•ด ์ฃผ์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค.
์ฒซ ๋ฒˆ์งธ ์งˆ๋ฌธ์€ ๋ชจ๊ฑด ์Šคํƒ ๋ฆฌ์—์„œ ํ•ด์ฃผ์…จ์Šต๋‹ˆ๋‹ค.
Unknown Analyst: So you aim to become an AI platform company, but so do AI and OpenAI. So how do you define an AI era platform company? Why do you think a start-up like MiniMax can become one?**Unknown Analyst:** AI ํ”Œ๋žซํผ ๊ธฐ์—…์ด ๋˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•˜์‹œ๋Š”๊ตฐ์š”. ๊ทธ๋Ÿฐ๋ฐ ๋‹ค๋ฅธ ์ฃผ์š” AI ๊ธฐ์—…๋“ค๊ณผ ์˜คํ”ˆAI(OpenAI)๋„ ๋งˆ์ฐฌ๊ฐ€์ง€์ธ๋ฐ์š”. ๊ทธ๋ ‡๋‹ค๋ฉด AI ์‹œ๋Œ€์˜ ํ”Œ๋žซํผ ๊ธฐ์—…์„ ์–ด๋–ป๊ฒŒ ์ •์˜ํ•˜์‹œ๋‚˜์š”? ๊ทธ๋ฆฌ๊ณ  ๋ฏธ๋‹ˆ๋งฅ์Šค(MiniMax) ๊ฐ™์€ ์Šคํƒ€ํŠธ์—…(start-up)์ด ๊ทธ๋ ‡๊ฒŒ ๋  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ณด์‹œ๋Š” ์ด์œ ๋Š” ๋ฌด์—‡์ธ๊ฐ€์š”?
Junjie Yan: Founder, Executive Chairman, CTO & CEO Thank you for your question. This is something we have been discussing and thinking about internally for a long time. So like we mentioned earlier, when the boundary of intelligence is pushed forward, it creates many new scenarios, new customers and new users, forming a new ecosystem and generating new commercialization dividends, such as, for instance, in coding and video or image generation, there are companies that have already emerged. Why MiniMax have the opportunity to become a platform company for the AI era? I think there are several reasons. For one, the AI market is not a zero-sum market. The incremental market each year is larger than the existing stock. And it is also not a winner-takes-all market. As long as you have unique differentiated innovation, then you have your market fit. We believe that our -- over the next 2 to 3 years, our model R&D capabilities and infrastructure capabilities are highly likely to create new scenarios, and there's tremendous innovation in market space in areas such as coding, office productivity and interactive entertainment. And in such a high growing -- fast-growing market, we feel like the opportunity lies in 3 areas or 3 layers. First is on the model layer. I think a critical element is that we rely on the long-term accumulation of model and faster iteration. For instance, over the 180 days, we successfully released M2, M2.1, M2.5, each bringing rapid growth in user numbers and API calls. And also, since day 1, we have been accumulating capabilities in cross modality. So we are the only company who has adopted this strategy, which position us advantageously in this inevitable trend of multimodal fusion. The second is on the product layer. MiniMax is the first domestic company who focus both on product and model. So the model plus product forms a stronger barrier to entry. So the model as a product is something hard to replicate by other peers. And the third layer is on the ecosystem side. We have leveraged our differentiated capability and have created an open system, for example, in open cloud. OpenClaw used many of our models to develop. And also is very fit for large thoroughput product and also using it to further integrate it. So this also further break down or reduce the barrier to entry for our users. So that's why we are seeing a lot of code contribution. So we can help -- we have the ability to help the ecosystem grow at rapid speed. And looking ahead, this is just the beginning of our own internal ecosystem or ecosystem we are building. Going forward, we are going to focus on building our next-generation M3 series of full-modality model, establishing clear model differentiation. On the other hand, we hope to build a distinctive products and ecosystem around the intelligence that we offer. We believe that alongside the major tech incumbents, we are the only company capable of executing on both product and models at the same time, even in Asia. Thank you.**Junjie Yan:** ์งˆ๋ฌธ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ถ€๋ถ„์€ ์ €ํฌ๊ฐ€ ๋‚ด๋ถ€์ ์œผ๋กœ ์˜ค๋žซ๋™์•ˆ ๋…ผ์˜ํ•˜๊ณ  ๊ณ ๋ฏผํ•ด ์˜จ ๋‚ด์šฉ์ž…๋‹ˆ๋‹ค. ์•ž์„œ ๋ง์”€๋“œ๋ฆฐ ๊ฒƒ์ฒ˜๋Ÿผ, ์ง€๋Šฅ์˜ ๊ฒฝ๊ณ„๊ฐ€ ํ™•์žฅ๋  ๋•Œ๋งˆ๋‹ค ๋งŽ์€ ์ƒˆ๋กœ์šด ์‹œ๋‚˜๋ฆฌ์˜ค, ์ƒˆ๋กœ์šด ๊ณ ๊ฐ, ์ƒˆ๋กœ์šด ์‚ฌ์šฉ์ž๋ฅผ ์ฐฝ์ถœํ•˜๊ณ  ์ƒˆ๋กœ์šด ์ƒํƒœ๊ณ„๋ฅผ ํ˜•์„ฑํ•˜๋ฉฐ ์ƒˆ๋กœ์šด ์ƒ์—…ํ™” ๋ฐฐ๋‹น๊ธˆ(commercialization dividends)์„ ๋งŒ๋“ค์–ด๋ƒ…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ฝ”๋”ฉ, ๋น„๋””์˜ค ๋˜๋Š” ์ด๋ฏธ์ง€ ์ƒ์„ฑ ๋ถ„์•ผ์—์„œ ์ด๋ฏธ ๋งŽ์€ ๊ธฐ์—…๋“ค์ด ๋“ฑ์žฅํ–ˆ์Šต๋‹ˆ๋‹ค. ์™œ MiniMax๊ฐ€ AI ์‹œ๋Œ€์˜ ํ”Œ๋žซํผ ๊ธฐ์—…์ด ๋  ๊ธฐํšŒ๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ๋Š”๊ฐ€? ์ €๋Š” ๋ช‡ ๊ฐ€์ง€ ์ด์œ ๊ฐ€ ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ฒซ์งธ๋กœ, AI ์‹œ์žฅ์€ ์ œ๋กœ์„ฌ ์‹œ์žฅ(zero-sum market)์ด ์•„๋‹™๋‹ˆ๋‹ค. ๋งค๋…„ ์ฆ๊ฐ€ํ•˜๋Š” ์‹œ์žฅ ๊ทœ๋ชจ๊ฐ€ ๊ธฐ์กด ์‹œ์žฅ ๊ทœ๋ชจ๋ณด๋‹ค ๋” ํฝ๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์Šน์ž๋…์‹ ์‹œ์žฅ(winner-takes-all market)๋„ ์•„๋‹™๋‹ˆ๋‹ค. ๋…์ ์ ์ธ ์ฐจ๋ณ„ํ™”๋œ ํ˜์‹ ์ด ์žˆ๋‹ค๋ฉด, ์‹œ์žฅ ์ ํ•ฉ์„ฑ(market fit)์„ ํ™•๋ณดํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ํ–ฅํ›„ 2~3๋…„๊ฐ„ ์ €ํฌ์˜ ๋ชจ๋ธ R&D ์—ญ๋Ÿ‰๊ณผ ์ธํ”„๋ผ ์—ญ๋Ÿ‰์ด ์ƒˆ๋กœ์šด ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ์ฐฝ์ถœํ•  ๊ฐ€๋Šฅ์„ฑ์ด ๋งค์šฐ ๋†’๋‹ค๊ณ  ๋ณด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ์ฝ”๋”ฉ, ์‚ฌ๋ฌด ์ƒ์‚ฐ์„ฑ(office productivity), ์ธํ„ฐ๋ž™ํ‹ฐ๋ธŒ ์—”ํ„ฐํ…Œ์ธ๋จผํŠธ(interactive entertainment)์™€ ๊ฐ™์€ ๋ถ„์•ผ์˜ ์‹œ์žฅ์—์„œ๋Š” ์—„์ฒญ๋‚œ ํ˜์‹ ์ด ์ผ์–ด๋‚˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์ด์ฒ˜๋Ÿผ ๋น ๋ฅด๊ฒŒ ์„ฑ์žฅํ•˜๋Š” ์‹œ์žฅ์—์„œ ๊ธฐํšŒ๋Š” ์„ธ ๊ฐ€์ง€ ์˜์—ญ, ํ˜น์€ ์„ธ ๊ฐ€์ง€ ๊ณ„์ธต(layer)์— ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ฒซ์งธ๋Š” ๋ชจ๋ธ ๊ณ„์ธต(model layer)์ž…๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ ํ•ต์‹ฌ ์š”์†Œ๋Š” ๋ชจ๋ธ์˜ ์žฅ๊ธฐ์ ์ธ ์ถ•์ ๊ณผ ๋น ๋ฅธ ๋ฐ˜๋ณต(iteration)์— ์˜์กดํ•œ๋‹ค๋Š” ์ ์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์ง€๋‚œ 180์ผ ๋™์•ˆ ์ €ํฌ๋Š” M2, M2.1, M2.5 ๋ฒ„์ „์„ ์„ฑ๊ณต์ ์œผ๋กœ ์ถœ์‹œํ–ˆ์œผ๋ฉฐ, ๊ฐ ๋ฒ„์ „์€ ์‚ฌ์šฉ์ž ์ˆ˜์™€ API ํ˜ธ์ถœ(API call) ์ˆ˜์˜ ๊ธ‰๊ฒฉํ•œ ์„ฑ์žฅ์„ ๊ฐ€์ ธ์™”์Šต๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ฒ˜์Œ๋ถ€ํ„ฐ ๊ต์ฐจ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(cross-modality) ๋ถ„์•ผ์—์„œ ์—ญ๋Ÿ‰์„ ๊พธ์ค€ํžˆ ์ถ•์ ํ•ด์™”์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ „๋žต์„ ์ฑ„ํƒํ•œ ์œ ์ผํ•œ ํšŒ์‚ฌ๋กœ์„œ, ์ด๋Š” ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์œตํ•ฉ(multimodal fusion)์ด๋ผ๋Š” ๊ฑฐ์Šค๋ฅผ ์ˆ˜ ์—†๋Š” ํ๋ฆ„ ์†์—์„œ ์ €ํฌ๋ฅผ ๋งค์šฐ ์œ ๋ฆฌํ•œ ์œ„์น˜์— ๋†“์ด๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.

๋‘ ๋ฒˆ์งธ๋Š” ์ œํ’ˆ ์ธก๋ฉด์ž…๋‹ˆ๋‹ค. MiniMax๋Š” ์ œํ’ˆ๊ณผ ๋ชจ๋ธ ๋ชจ๋‘์— ์ง‘์ค‘ํ•˜๋Š” ๊ตญ๋‚ด ์ตœ์ดˆ์˜ ํšŒ์‚ฌ์ž…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ชจ๋ธ๊ณผ ์ œํ’ˆ์˜ ๊ฒฐํ•ฉ์€ ๋”์šฑ ๊ฐ•๋ ฅํ•œ ์ง„์ž… ์žฅ๋ฒฝ(barrier to entry)์„ ํ˜•์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰, ์ œํ’ˆ์œผ๋กœ์„œ์˜ ๋ชจ๋ธ์€ ๋‹ค๋ฅธ ๊ฒฝ์Ÿ์‚ฌ๋“ค์ด ์‰ฝ๊ฒŒ ๋ณต์ œํ•˜๊ธฐ ์–ด๋ ค์šด ๊ฐ•์ ์ž…๋‹ˆ๋‹ค.

์„ธ ๋ฒˆ์งธ๋Š” ์ƒํƒœ๊ณ„(ecosystem) ์ธก๋ฉด์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ฐจ๋ณ„ํ™”๋œ ์—ญ๋Ÿ‰์„ ํ™œ์šฉํ•˜์—ฌ ๊ฐœ๋ฐฉํ˜• ์‹œ์Šคํ…œ(open system)์„ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์˜คํ”ˆ ํด๋ผ์šฐ๋“œ(open cloud) ๋ถ„์•ผ๊ฐ€ ๊ทธ๋ ‡์Šต๋‹ˆ๋‹ค. OpenClaw๋Š” ์ €ํฌ์˜ ๋งŽ์€ ๋ชจ๋ธ๋“ค์„ ํ™œ์šฉํ•˜์—ฌ ๊ฐœ๋ฐœ๋˜์—ˆ์œผ๋ฉฐ, ๋Œ€๊ทœ๋ชจ ์ฒ˜๋ฆฌ๋Ÿ‰(large throughput) ์ œํ’ˆ์— ๋งค์šฐ ์ ํ•ฉํ•˜๊ณ , ์ด๋ฅผ ํ†ตํ•ด ์ถ”๊ฐ€์ ์ธ ํ†ตํ•ฉ์„ ๊ณ„์†ํ•ด์„œ ์ด๋ฃจ์–ด๋‚ด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋˜ํ•œ ์ €ํฌ ์‚ฌ์šฉ์ž๋“ค์˜ ์ง„์ž… ์žฅ๋ฒฝ(barrier to entry)์„ ๋”์šฑ ๋‚ฎ์ถ”๊ฑฐ๋‚˜ ํ—ˆ๋ฌด๋Š” ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ ‡๊ธฐ ๋•Œ๋ฌธ์— ์ €ํฌ๋Š” ๋งŽ์€ ์ฝ”๋“œ ๊ธฐ์—ฌ(code contribution)๋ฅผ ๋ณด๊ณ  ์žˆ์œผ๋ฉฐ, ์ƒํƒœ๊ณ„(ecosystem)๊ฐ€ ๋น ๋ฅธ ์†๋„๋กœ ์„ฑ์žฅํ•  ์ˆ˜ ์žˆ๋„๋ก ๋„์šธ ์—ญ๋Ÿ‰์„ ๊ฐ–์ถ”๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์•ž์œผ๋กœ๋ฅผ ๋‚ด๋‹ค๋ณด๋ฉด, ์ด๋Š” ์ €ํฌ ์ž์ฒด ๋‚ด๋ถ€ ์ƒํƒœ๊ณ„, ์ฆ‰ ์ €ํฌ๊ฐ€ ๊ตฌ์ถ•ํ•˜๊ณ  ์žˆ๋Š” ์ƒํƒœ๊ณ„์˜ ์‹œ์ž‘์— ๋ถˆ๊ณผํ•ฉ๋‹ˆ๋‹ค. ํ–ฅํ›„ ์ €ํฌ๋Š” ์ฐจ์„ธ๋Œ€ M3 ์‹œ๋ฆฌ์ฆˆ ํ’€ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ๋ชจ๋ธ(full-modality model) ๊ตฌ์ถ•์— ์ง‘์ค‘ํ•˜๊ณ , ๋ช…ํ™•ํ•œ ๋ชจ๋ธ ์ฐจ๋ณ„ํ™”(model differentiation)๋ฅผ ํ™•๋ฆฝํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํ•œํŽธ์œผ๋กœ๋Š” ์ €ํฌ๊ฐ€ ์ œ๊ณตํ•˜๋Š” ์ธํ…”๋ฆฌ์ „์Šค(intelligence)๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ์ฐจ๋ณ„ํ™”๋œ ์ œํ’ˆ๊ณผ ์ƒํƒœ๊ณ„๋ฅผ ๊ตฌ์ถ•ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ฃผ์š” ๊ธฐ์ˆ  ์„ ๋„ ๊ธฐ์—…(tech incumbents)๋“ค๊ณผ ๋”๋ถˆ์–ด, ์•„์‹œ์•„์—์„œ ์ œํ’ˆ๊ณผ ๋ชจ๋ธ์„ ๋™์‹œ์— ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ๋Š” ์œ ์ผํ•œ ๊ธฐ์—…์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
Operator: Next question comes from Alex Yao of JPMorgan.**Operator:** ๋‹ค์Œ ์งˆ๋ฌธ์€ JP๋ชจ๊ฑด(JPMorgan)์˜ ์•Œ๋ ‰์Šค ์•ผ์˜ค ๋‹˜์ž…๋‹ˆ๋‹ค.
Alex Yao: JPMorgan Chase & Co, Research Division Congratulations on the strong results. I want to ask about multi-modality, which is something that you emphasize as the end game. So if competitors focus on perfecting a single modality first and then switch to cross-modality, means that might they move faster than you? So your approach of focusing on cross-modality in the first place will be burdensome for you? Will that be the case?**Alex Yao:** ๋จผ์ €, ์ข‹์€ ์‹ค์ ์„ ๊ฑฐ๋‘์‹  ๊ฒƒ์„ ์ถ•ํ•˜๋“œ๋ฆฝ๋‹ˆ๋‹ค.

์ตœ์ข… ๋ชฉํ‘œ(end game)๋กœ ๊ฐ•์กฐํ•˜์‹œ๋Š” ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(multi-modality)์— ๋Œ€ํ•ด ์งˆ๋ฌธ๋“œ๋ฆฌ๊ณ  ์‹ถ์Šต๋‹ˆ๋‹ค. ๋งŒ์•ฝ ๊ฒฝ์Ÿ์‚ฌ๋“ค์ด ๋จผ์ € ๋‹จ์ผ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(single modality)๋ฅผ ์™„๋ฒฝํ•˜๊ฒŒ ๊ตฌ์ถ•ํ•œ ๋‹ค์Œ ๊ต์ฐจ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(cross-modality)๋กœ ์ „ํ™˜ํ•˜๋Š” ์ „๋žต์„ ํƒํ•œ๋‹ค๋ฉด, ์˜คํžˆ๋ ค ์ €ํฌ๋ณด๋‹ค ๋” ๋น ๋ฅด๊ฒŒ ์‹œ์žฅ์„ ์„ ์ ํ•  ์ˆ˜๋„ ์žˆ์ง€ ์•Š์„๊นŒ์š”? ๊ทธ๋ ‡๋‹ค๋ฉด ์ฒ˜์Œ๋ถ€ํ„ฐ ๊ต์ฐจ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ์— ์ง‘์ค‘ํ•˜์‹œ๋Š” ์—ฌ๋Ÿฌ๋ถ„์˜ ์ ‘๊ทผ ๋ฐฉ์‹์ด ์˜คํžˆ๋ ค ๋ถ€๋‹ด์œผ๋กœ ์ž‘์šฉํ•  ์ˆ˜๋„ ์žˆ๋‹ค๋Š” ์˜๋ฏธ์ธ์ง€์š”? ์–ด๋–ป๊ฒŒ ์ƒ๊ฐํ•˜์‹œ๋Š”์ง€ ๊ถ๊ธˆํ•ฉ๋‹ˆ๋‹ค.
Junjie Yan: Founder, Executive Chairman, CTO & CEO Thank you for the question. This is something we have been asked since the first day when the company was founded. I would like to take this opportunity why we focus on cross-modality. We believe that the integration of modeling modality is the fundamental prerequisite for continuously improving intelligence. Over the past 6 months, several models have validated this trend by achieving breakthroughs through multimodal integration. For example, models like Nano and Nano Pro integrate visual understanding and generation, further expanding intelligence boundaries. For us, we -- it's 2-stage approach for us. We are now in the second stage. Over the past 4 years, which is Stage 1 for us, we have steadily built industry-leading models in each modality, creating strong positive influence and market recognition. There are many models that we are offering across modalities, and we have made quite a significant achievements in respective field. And next up, the critical thing is to integrate them and fuse them to make greater breakthroughs. In the second half of this year, the M3 models is designed to achieve that goal. And along this, approach, we want to emphasize that accumulation in each modality is a long-term process. It takes time from data to single modality and then to multimodal integration. The entire chain requires significant time. So this is the foundation of our long-term capabilities and what sets us apart. We are one of the only three companies in China that have achieved the leadership across every modality. And the second point I want to share is that for video generation, other than coding and agentic tasks, is the largest market. We believe that we are able to see mid- to long-term form content near real-time formats. We believe that we are -- we can also achieve that kind of ability, and there is a significant opportunity out there for us. And like you mentioned, will our strategic approach hinder our R&D development? Well, how do we put it? I think there are challenges, but they are inevitable. Since our founding, AGI has been always multimodal input and output. Therefore, we have built an organization structure that enables reusing foundational capabilities across the modalities. As you can see, under this AI-native organizational architecture, our cost of building full modality is not higher than that of other start-ups and is far lower than the investment of large tech companies. So moreover, each individual modality has achieved a competitive model. In some cases, outperforming companies focus solely on a single modality. So our technical judgment and forward-looking positioning have been continuously validated over the past few years and will only become clearer going forward. Thank you.**Junjie Yan:** ์งˆ๋ฌธํ•ด์ฃผ์…”์„œ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ์งˆ๋ฌธ์€ ์ €ํฌ ํšŒ์‚ฌ๊ฐ€ ์„ค๋ฆฝ๋œ ์ฒซ๋‚ ๋ถ€ํ„ฐ ๊ณ„์† ๋ฐ›์•„์˜จ ์งˆ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ด ๊ธฐํšŒ๋ฅผ ๋นŒ์–ด ์ €ํฌ๊ฐ€ ์™œ ํฌ๋กœ์Šค ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(cross-modality)์— ์ง‘์ค‘ํ•˜๋Š”์ง€ ๋ง์”€๋“œ๋ฆฌ๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ๋ชจ๋ธ๋ง ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(modeling modality)์˜ ํ†ตํ•ฉ์ด ์ง€๋Šฅ์„ ์ง€์†์ ์œผ๋กœ ํ–ฅ์ƒ์‹œํ‚ค๋Š” ๋ฐ ์žˆ์–ด ๊ทผ๋ณธ์ ์ธ ์ „์ œ ์กฐ๊ฑด์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ์ง€๋‚œ 6๊ฐœ์›” ๋™์•ˆ, ์—ฌ๋Ÿฌ ๋ชจ๋ธ๋“ค์ด ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ํ†ตํ•ฉ(multimodal integration)์„ ํ†ตํ•ด ํš๊ธฐ์ ์ธ ๋ฐœ์ „์„ ์ด๋ฃจ๋ฉฐ ์ด๋Ÿฌํ•œ ์ถ”์„ธ๋ฅผ ์ž…์ฆํ–ˆ์Šต๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, Nano์™€ Nano Pro ๊ฐ™์€ ๋ชจ๋ธ๋“ค์€ ์‹œ๊ฐ์  ์ดํ•ด์™€ ์ƒ์„ฑ์„ ํ†ตํ•ฉํ•˜์—ฌ ์ง€๋Šฅ์˜ ๊ฒฝ๊ณ„๋ฅผ ๋”์šฑ ํ™•์žฅํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์ €ํฌ์˜ ์ ‘๊ทผ ๋ฐฉ์‹์€ ๋‘ ๋‹จ๊ณ„๋กœ ์ด๋ฃจ์–ด์ ธ ์žˆ์œผ๋ฉฐ, ํ˜„์žฌ ์ €ํฌ๋Š” ๋‘ ๋ฒˆ์งธ ๋‹จ๊ณ„์— ์žˆ์Šต๋‹ˆ๋‹ค. ์ง€๋‚œ 4๋…„๊ฐ„, ์ €ํฌ์—๊ฒŒ๋Š” 1๋‹จ๊ณ„(Stage 1)์˜€์Šต๋‹ˆ๋‹ค๋งŒ, ๊ฐ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(modality)๋ณ„๋กœ ์—…๊ณ„ ์„ ๋„์ ์ธ ๋ชจ๋ธ๋“ค์„ ๊พธ์ค€ํžˆ ๊ตฌ์ถ•ํ•ด์™”๊ณ , ์ด๋ฅผ ํ†ตํ•ด ๊ฐ•๋ ฅํ•œ ๊ธ์ •์  ์˜ํ–ฅ๋ ฅ๊ณผ ์‹œ์žฅ ์ธ์ง€๋„๋ฅผ ๋งŒ๋“ค์–ด๋ƒˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ๊ฐ€ ์ œ๊ณตํ•˜๋Š” ๋ชจ๋ธ๋“ค์€ ๋‹ค์–‘ํ•œ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ์— ๊ฑธ์ณ ๋งŽ์ด ์žˆ์œผ๋ฉฐ, ๊ฐ ๋ถ„์•ผ์—์„œ ์ƒ๋‹นํ•œ ์„ฑ๊ณผ๋ฅผ ๊ฑฐ๋‘์—ˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ๋‹ค์Œ์œผ๋กœ ํ•ต์‹ฌ์€ ์ด๋“ค์„ ํ†ตํ•ฉํ•˜๊ณ  ์œตํ•ฉํ•˜์—ฌ ๋” ํฐ ๋ŒํŒŒ๊ตฌ๋ฅผ ๋งˆ๋ จํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์˜ฌํ•ด ํ•˜๋ฐ˜๊ธฐ์—๋Š” M3 ๋ชจ๋ธ๋“ค์ด ๋ฐ”๋กœ ๊ทธ ๋ชฉํ‘œ๋ฅผ ๋‹ฌ์„ฑํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

์ด๋Ÿฌํ•œ ์ ‘๊ทผ ๋ฐฉ์‹๊ณผ ๋”๋ถˆ์–ด, ๊ฐ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ์—์„œ์˜ ์ถ•์ ์€ ์žฅ๊ธฐ์ ์ธ ๊ณผ์ •์ด๋ผ๋Š” ์ ์„ ๊ฐ•์กฐํ•˜๊ณ  ์‹ถ์Šต๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ(data)์—์„œ ๋‹จ์ผ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(single modality)๋กœ, ๊ทธ๋ฆฌ๊ณ  ๋‹ค์‹œ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ(multimodal) ํ†ตํ•ฉ์œผ๋กœ ๋‚˜์•„๊ฐ€๋Š” ๋ฐ๊นŒ์ง€๋Š” ์‹œ๊ฐ„์ด ๊ฑธ๋ฆฝ๋‹ˆ๋‹ค. ์ด ์ „์ฒด ๊ณผ์ •์€ ์ƒ๋‹นํ•œ ์‹œ๊ฐ„์„ ํ•„์š”๋กœ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด๊ฒƒ์ด ์ €ํฌ์˜ ์žฅ๊ธฐ์ ์ธ ์—ญ๋Ÿ‰์˜ ๊ธฐ๋ฐ˜์ด๋ฉฐ, ์ €ํฌ๋ฅผ ์ฐจ๋ณ„ํ™”ํ•˜๋Š” ์š”์†Œ์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ค‘๊ตญ ๋‚ด์—์„œ ๋ชจ๋“  ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(modality) ๋ถ„์•ผ์—์„œ ์„ ๋‘๋ฅผ ๋‹ฌ์„ฑํ•œ ์œ ์ผํ•œ ์„ธ ๊ฐœ ํšŒ์‚ฌ ์ค‘ ํ•˜๋‚˜์ž…๋‹ˆ๋‹ค. ๋‘ ๋ฒˆ์งธ๋กœ ๋ง์”€๋“œ๋ฆฌ๊ณ  ์‹ถ์€ ์ ์€, ๋น„๋””์˜ค ์ƒ์„ฑ(video generation) ๋ถ„์•ผ๊ฐ€ ์ฝ”๋”ฉ ๋ฐ ์—์ด์ „ํŠธ(agentic) ์ž‘์—… ์™ธ์— ๊ฐ€์žฅ ํฐ ์‹œ์žฅ์ด๋ผ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ค‘์žฅ๊ธฐ ํ˜•ํƒœ์˜ ์ฝ˜ํ…์ธ ๋ฅผ ๊ฑฐ์˜ ์‹ค์‹œ๊ฐ„(real-time) ํ˜•์‹์œผ๋กœ ์ƒ์„ฑํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ๋ฏฟ์œผ๋ฉฐ, ๊ทธ๋Ÿฌํ•œ ์—ญ๋Ÿ‰(ability) ๋˜ํ•œ ๋‹ฌ์„ฑํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ์—๊ฒŒ๋Š” ์ƒ๋‹นํ•œ ๊ธฐํšŒ(opportunity)๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ๋ง์”€ํ•˜์…จ๋“ฏ์ด, ์ €ํฌ์˜ ์ „๋žต์  ์ ‘๊ทผ ๋ฐฉ์‹(strategic approach)์ด R&D ๊ฐœ๋ฐœ(R&D development)์„ ์ €ํ•ดํ• ์ง€์— ๋Œ€ํ•œ ์งˆ๋ฌธ์— ๋Œ€ํ•ด ๋ง์”€๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค. ์Œ, ์–ด๋–ป๊ฒŒ ๋ง์”€๋“œ๋ ค์•ผ ํ• ๊นŒ์š”? ๋ฌผ๋ก  ์–ด๋ ค์›€(challenges)์ด ์žˆ๊ฒ ์ง€๋งŒ, ์ด๋Š” ํ”ผํ•  ์ˆ˜ ์—†๋Š” ๋ถ€๋ถ„์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ฐฝ๋ฆฝ ์ด๋ž˜๋กœ AGI๊ฐ€ ํ•ญ์ƒ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ(multimodal) ์ž…์ถœ๋ ฅ(input and output)์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•ด์•ผ ํ•œ๋‹ค๊ณ  ๋ณด์•„์™”์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ €ํฌ๋Š” ๋ชจ๋“  ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(modality) ์ „๋ฐ˜์— ๊ฑธ์ณ ๊ธฐ๋ฐ˜ ์—ญ๋Ÿ‰(foundational capabilities)์„ ์žฌ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•˜๋Š” ์กฐ์ง ๊ตฌ์กฐ(organization structure)๋ฅผ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ณด์‹œ๋‹ค์‹œํ”ผ, ์ด๋Ÿฌํ•œ AI ๋„ค์ดํ‹ฐ๋ธŒ(AI-native) ์กฐ์ง ์•„ํ‚คํ…์ฒ˜(organizational architecture) ํ•˜์—์„œ ์ €ํฌ์˜ ํ’€ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(full modality) ๊ตฌ์ถ• ๋น„์šฉ์€ ๋‹ค๋ฅธ ์Šคํƒ€ํŠธ์—…(start-up)๋“ค๊ณผ ๋น„๊ตํ–ˆ์„ ๋•Œ ๋” ๋†’์ง€ ์•Š์œผ๋ฉฐ, ๋Œ€ํ˜• ๊ธฐ์ˆ  ๊ธฐ์—…(large tech companies)๋“ค์˜ ํˆฌ์ž์•ก๋ณด๋‹ค ํ›จ์”ฌ ๋‚ฎ์Šต๋‹ˆ๋‹ค.

๋˜ํ•œ, ๊ฐ๊ฐ์˜ ๊ฐœ๋ณ„ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ์—ญ์‹œ ๊ฒฝ์Ÿ๋ ฅ ์žˆ๋Š” ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ผ๋ถ€ ๊ฒฝ์šฐ์—๋Š” ๋‹จ์ผ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(single modality)์—๋งŒ ์ง‘์ค‘ํ•˜๋Š” ๊ธฐ์—…๋“ค๋ณด๋‹ค๋„ ๋›ฐ์–ด๋‚œ ์„ฑ๊ณผ๋ฅผ ๋‚ด๊ธฐ๋„ ํ–ˆ์Šต๋‹ˆ๋‹ค.

๋”ฐ๋ผ์„œ ์ €ํฌ์˜ ๊ธฐ์ˆ ์  ํŒ๋‹จ๊ณผ ๋ฏธ๋ž˜ ์ง€ํ–ฅ์ ์ธ ํฌ์ง€์…”๋‹(positioning)์€ ์ง€๋‚œ ๋ช‡ ๋…„๊ฐ„ ์ง€์†์ ์œผ๋กœ ๊ฒ€์ฆ๋˜์–ด ์™”์œผ๋ฉฐ, ์•ž์œผ๋กœ๋Š” ๋”์šฑ ๋ช…ํ™•ํ•ด์งˆ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
Operator: Your next question comes from UBS, Kenny Fong.**Operator:** ๋‹ค์Œ ์งˆ๋ฌธ์€ UBS์˜ ์ผ€๋‹ˆ ํ ๋‹˜๊ป˜์„œ ํ•ด์ฃผ์‹œ๊ฒ ์Šต๋‹ˆ๋‹ค.
Kenneth Fong: UBS Investment Bank, Research Division Congratulations on the strong results following your IPO. You mentioned that L4 to L5 levels of programming intelligence are approaching, and there are many claims that many software companies may be replaced by agents. How should we view this transformation? And what is your position within it?**Kenneth Fong:** IPO(๊ธฐ์—…๊ณต๊ฐœ) ์ดํ›„ ๊ฒฌ์กฐํ•œ ์‹ค์ ์„ ๋‚ด์‹  ๊ฒƒ์— ๋Œ€ํ•ด ์ถ•ํ•˜๋“œ๋ฆฝ๋‹ˆ๋‹ค.

L4์—์„œ L5 ์ˆ˜์ค€์˜ ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์ง€๋Šฅ(programming intelligence)์ด ๋‹ค๊ฐ€์˜ค๊ณ  ์žˆ๋‹ค๊ณ  ๋ง์”€ํ•˜์…จ๋Š”๋ฐ, ๋งŽ์€ ์†Œํ”„ํŠธ์›จ์–ด ํšŒ์‚ฌ๋“ค์ด ์—์ด์ „ํŠธ(agent)๋กœ ๋Œ€์ฒด๋  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ฃผ์žฅ์ด ๋งŽ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ณ€ํ™”๋ฅผ ์–ด๋–ป๊ฒŒ ๋ณด์•„์•ผ ํ• ๊นŒ์š”? ๊ทธ๋ฆฌ๊ณ  ์ด ๋ณ€ํ™” ์†์—์„œ ๊ท€์‚ฌ์˜ ์ž…์ง€๋Š” ์–ด๋–ป์Šต๋‹ˆ๊นŒ?
Junjie Yan: Founder, Executive Chairman, CTO & CEO Well, this is a very important question. Let me first explain what L4 to L5 levels of intelligence mean and the future direction of programming intelligence and where we are within the transformation. For L3, these are just agents we're using today, while L4, L5 represents colleague level and organizational level intelligence. To give you an example -- we want to build world-leading models. It takes collaboration of many people, innovation and experiment in algorithms, optimization of programs, processing of data and tech ops. Lots of work. We feel that L4 will be able to handle a lot of innovative tasks, for example, conducting experiments based on a research paper and come up with efficient solutions to many of the challenges in the research paper. So a lot of innovations. For L5 level of intelligence, it takes not only one person, but also a collaboration of many people, many employees. I think coding is only a part of agents. It was the earliest productivity capability validated. Other than that, we believe that office productivity will replicate last year's rapid progress in coding over the next year. We believe the market is even bigger than coding. And with that being said, how do we see ourselves? How do we position ourselves? I think we are having a huge market in front of us. Coding models allow more people to write code but to code it even better. But again, I would like to emphasize that coders remain a small portion of the labor market. A much larger portion of the workplace now is handled by non-code software. Use cases such as data analysis as well as financial modeling or presentation slides that used to support a financial results conference. So these use cases, work use cases represent a far larger market than coding. We have already achieved early progress in coding and agents, securing a unique market position with minimal resources. So the bigger market is just -- penetration into the bigger market is just the beginning. And for us, we move fast. Like I said, the evolution from M2 all the way to M2.3 takes only 100 days. So literally, we maintained the fastest iteration speed in the industry with each generation achieving significant improvements in both capability and usage. That underscores our R&D capability and our ability to handle scale. So we build M2 with limited resources, but our resources are scaling up. I believe, with that, model improvement will accelerate, and better models will raise the ceiling further. So our historical performance was built on the M2 series models. And we expect the M3 model series will unlock even greater potential, creating a positive flywheel effect. Other than our fast move, we are able to create differentiated models, which has been repeatedly validated over the past few months. Like I said, the market is huge and technical path will diverge. For us, we need to know we have -- whether we have the ability to define technical road maps. We do not aim to win across every dimension. Instead, we focus on defining model capabilities that showcase our distinct strength. For M2, Hailuo 2 and Speech 2 series model, each established a clear differentiation, and we're able to gain rapid market traction. So its characteristics by low latency, high cost efficiencies. So these characteristics set us apart and helping us to gain bigger market share. As our organization and resources continue to scale, our deep understanding of model evolution and technical road maps will further strengthen this differentiation and its value. In summary, we are confident in further increasing our share and achieving more breakthroughs in coding through the agents and the broader productivity market. So we hope to make greater breakthroughs in gaining a bigger market share with our faster iteration and a stronger differentiation position. Thank you.**Junjie Yan:** ์Œ, ๋งค์šฐ ์ค‘์š”ํ•œ ์งˆ๋ฌธ์ž…๋‹ˆ๋‹ค.

๋จผ์ € L4์—์„œ L5 ์ˆ˜์ค€์˜ ์ง€๋Šฅ(intelligence)์ด ๋ฌด์—‡์„ ์˜๋ฏธํ•˜๋Š”์ง€, ํ”„๋กœ๊ทธ๋ž˜๋ฐ ์ง€๋Šฅ์˜ ๋ฏธ๋ž˜ ๋ฐฉํ–ฅ์€ ์–ด๋– ํ•œ์ง€, ๊ทธ๋ฆฌ๊ณ  ์ด๋Ÿฌํ•œ ์ „ํ™˜ ๊ณผ์ •์—์„œ ์ €ํฌ๊ฐ€ ์–ด๋””์— ์œ„์น˜ํ•ด ์žˆ๋Š”์ง€ ์„ค๋ช…ํ•ด ๋“œ๋ฆฌ๊ฒ ์Šต๋‹ˆ๋‹ค.

L3๋Š” ์˜ค๋Š˜๋‚  ์šฐ๋ฆฌ๊ฐ€ ์‚ฌ์šฉํ•˜๋Š” ์—์ด์ „ํŠธ(agent)์— ๋ถˆ๊ณผํ•˜์ง€๋งŒ, L4์™€ L5๋Š” ๋™๋ฃŒ ์ˆ˜์ค€(colleague level) ๋ฐ ์กฐ์ง ์ˆ˜์ค€(organizational level)์˜ ์ง€๋Šฅ์„ ๋‚˜ํƒ€๋ƒ…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด ์„ค๋ช…ํ•ด ๋“œ๋ฆฌ์ž๋ฉด, ์ €ํฌ๋Š” ์„ธ๊ณ„ ์ตœ๊ณ  ์ˆ˜์ค€์˜ ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด์„œ๋Š” ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์˜ ํ˜‘์—…์ด ํ•„์š”ํ•˜๊ณ , ์•Œ๊ณ ๋ฆฌ์ฆ˜(algorithm) ํ˜์‹ ๊ณผ ์‹คํ—˜, ํ”„๋กœ๊ทธ๋žจ(program) ์ตœ์ ํ™”, ๋ฐ์ดํ„ฐ(data) ์ฒ˜๋ฆฌ ๋ฐ ๊ธฐ์ˆ  ์šด์˜(tech ops) ๋“ฑ ๋งŽ์€ ์ž‘์—…์ด ์ˆ˜๋ฐ˜๋ฉ๋‹ˆ๋‹ค.

์ €ํฌ๋Š” L4 ์ˆ˜์ค€์˜ ์ง€๋Šฅ์ด ๋งŽ์€ ํ˜์‹ ์ ์ธ ์ž‘์—…๋“ค์„ ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ์„ ๊ฒƒ์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์—ฐ๊ตฌ ๋…ผ๋ฌธ(research paper)์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์‹คํ—˜์„ ์ˆ˜ํ–‰ํ•˜๊ณ , ํ•ด๋‹น ๋…ผ๋ฌธ์˜ ์—ฌ๋Ÿฌ ๋‚œ์ œ์— ๋Œ€ํ•œ ํšจ์œจ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•˜๋Š” ๊ฒƒ๊ณผ ๊ฐ™์€ ์ผ๋“ค ๋ง์ž…๋‹ˆ๋‹ค. ๋งŽ์€ ํ˜์‹ ์ด ์ด๋ฃจ์–ด์ง€๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. L5 ์ˆ˜์ค€์˜ ์ธ๊ณต์ง€๋Šฅ(AI)์„ ์œ„ํ•ด์„œ๋Š” ํ•œ ์‚ฌ๋žŒ์˜ ๋…ธ๋ ฅ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, ๋งŽ์€ ์‚ฌ๋žŒ, ์ฆ‰ ๋งŽ์€ ์ง์›๋“ค์˜ ํ˜‘์—…์ด ํ•„์ˆ˜์ ์ž…๋‹ˆ๋‹ค. ์ €๋Š” ์ฝ”๋”ฉ(coding)์€ ์—์ด์ „ํŠธ(agents)๊ฐ€ ํ•˜๋Š” ์ผ์˜ ์ผ๋ถ€์ผ ๋ฟ์ด๋ผ๊ณ  ๋ด…๋‹ˆ๋‹ค. ์ฝ”๋”ฉ์€ ๊ฐ€์žฅ ๋จผ์ € ์ƒ์‚ฐ์„ฑ ์—ญ๋Ÿ‰(productivity capability)์ด ๊ฒ€์ฆ๋œ ๋ถ„์•ผ์˜€์Šต๋‹ˆ๋‹ค.

๊ทธ ์™ธ์—๋„, ์ €ํฌ๋Š” ๋‚ด๋…„์—๋Š” ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ ์ž‘๋…„์— ๋ณด์˜€๋˜ ๊ธ‰๊ฒฉํ•œ ๋ฐœ์ „์ด ์‚ฌ๋ฌด ์ƒ์‚ฐ์„ฑ(office productivity) ๋ถ„์•ผ์—์„œ๋„ ์žฌํ˜„๋  ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ์ด ์‹œ์žฅ์€ ์ฝ”๋”ฉ ์‹œ์žฅ๋ณด๋‹ค ํ›จ์”ฌ ๋” ํฌ๋‹ค๊ณ  ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ ‡๋‹ค๋ฉด ์šฐ๋ฆฌ๋Š” ์Šค์Šค๋กœ๋ฅผ ์–ด๋–ป๊ฒŒ ๋ณด๊ณ , ์–ด๋–ป๊ฒŒ ํฌ์ง€์…”๋‹(positioning)ํ•ด์•ผ ํ• ๊นŒ์š”?

์šฐ๋ฆฌ ์•ž์—๋Š” ์—„์ฒญ๋‚œ ์‹œ์žฅ ๊ธฐํšŒ๊ฐ€ ํŽผ์ณ์ ธ ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ฝ”๋”ฉ ๋ชจ๋ธ(coding models)์€ ๋” ๋งŽ์€ ์‚ฌ๋žŒ๋“ค์ด ์ฝ”๋“œ๋ฅผ ์ž‘์„ฑํ•˜๊ณ , ๋” ๋‚˜์€ ์ฝ”๋“œ๋ฅผ ๋งŒ๋“ค ์ˆ˜ ์žˆ๋„๋ก ํ•ด์ค๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ๋‹ค์‹œ ํ•œ๋ฒˆ ๊ฐ•์กฐํ•˜๊ณ  ์‹ถ์€ ์ ์€ ์ฝ”๋”(coder)๋“ค์€ ์—ฌ์ „ํžˆ ๋…ธ๋™ ์‹œ์žฅ์˜ ์ž‘์€ ๋ถ€๋ถ„์„ ์ฐจ์ง€ํ•œ๋‹ค๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ํ˜„์žฌ ์ง์žฅ์˜ ํ›จ์”ฌ ๋” ํฐ ๋ถ€๋ถ„์€ ๋น„(้ž)์ฝ”๋”ฉ ์†Œํ”„ํŠธ์›จ์–ด(non-code software)๋กœ ์ฒ˜๋ฆฌ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์  ๋ฐœํ‘œํšŒ(financial results conference)๋ฅผ ์ง€์›ํ•˜๋Š” ๋ฐ ์‚ฌ์šฉ๋˜์—ˆ๋˜ ๋ฐ์ดํ„ฐ ๋ถ„์„, ์žฌ๋ฌด ๋ชจ๋ธ๋ง(financial modeling), ๊ทธ๋ฆฌ๊ณ  ๋ฐœํ‘œ ์ž๋ฃŒ(presentation slides)์™€ ๊ฐ™์€ ํ™œ์šฉ ์‚ฌ๋ก€๋“ค์„ ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ์—…๋ฌด ํ™œ์šฉ ์‚ฌ๋ก€๋“ค์€ ์ฝ”๋”ฉ(coding) ์‹œ์žฅ๋ณด๋‹ค ํ›จ์”ฌ ๋” ํฐ ์‹œ์žฅ์„ ๋Œ€ํ‘œํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ด๋ฏธ ์ฝ”๋”ฉ ๋ฐ ์—์ด์ „ํŠธ(agent) ๋ถ„์•ผ์—์„œ ์ดˆ๊ธฐ ์„ฑ๊ณผ๋ฅผ ๋‹ฌ์„ฑํ•˜๋ฉฐ, ์ตœ์†Œํ•œ์˜ ์ž์›์œผ๋กœ ๋…๋ณด์ ์ธ ์‹œ์žฅ ์ง€์œ„๋ฅผ ํ™•๋ณดํ–ˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋” ํฐ ์‹œ์žฅ์œผ๋กœ์˜ ์ง„์ถœ์€ ์ด์ œ ๋ง‰ ์‹œ์ž‘์— ๋ถˆ๊ณผํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๋Š” ๋งค์šฐ ๋น ๋ฅด๊ฒŒ ์›€์ง์ž…๋‹ˆ๋‹ค. ์•ž์„œ ๋ง์”€๋“œ๋ ธ๋“ฏ์ด, M2๋ถ€ํ„ฐ M2.3๊นŒ์ง€ ์ง„ํ™”ํ•˜๋Š” ๋ฐ ๋‹จ 100์ผ๋ฐ–์— ๊ฑธ๋ฆฌ์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๋ง ๊ทธ๋Œ€๋กœ, ์ €ํฌ๋Š” ๊ฐ ์„ธ๋Œ€๊ฐ€ ์—ญ๋Ÿ‰๊ณผ ํ™œ์šฉ๋„ ๋ฉด์—์„œ ์ƒ๋‹นํ•œ ๊ฐœ์„ ์„ ์ด๋ฃจ์–ด๋‚ด๋ฉด์„œ ์—…๊ณ„์—์„œ ๊ฐ€์žฅ ๋น ๋ฅธ ์ดํ„ฐ๋ ˆ์ด์…˜ ์†๋„(iteration speed)๋ฅผ ์œ ์ง€ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ €ํฌ์˜ R&D ์—ญ๋Ÿ‰๊ณผ ํ™•์žฅ์„ฑ ์ฒ˜๋ฆฌ ๋Šฅ๋ ฅ(ability to handle scale)์„ ๋ช…ํ™•ํžˆ ์ž…์ฆํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ œํ•œ๋œ ์ž์›์œผ๋กœ M2 ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ–ˆ์ง€๋งŒ, ํ˜„์žฌ ์ €ํฌ์˜ ์ž์›์€ ํ™•๋Œ€๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ €๋Š” ์ด๋ฅผ ํ†ตํ•ด ๋ชจ๋ธ ๊ฐœ์„ ์ด ๊ฐ€์†ํ™”๋˜๊ณ , ๋” ๋‚˜์€ ๋ชจ๋ธ๋“ค์ด ์ž ์žฌ๋ ฅ์„ ๋”์šฑ ๋†’์ผ ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ์ €ํฌ์˜ ๊ณผ๊ฑฐ ์‹ค์ ์€ M2 ์‹œ๋ฆฌ์ฆˆ ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ๋‹ฌ์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  M3 ๋ชจ๋ธ ์‹œ๋ฆฌ์ฆˆ๋Š” ํ›จ์”ฌ ๋” ํฐ ์ž ์žฌ๋ ฅ(potential)์„ ๋ฐœํœ˜ํ•˜๊ณ  ๊ธ์ •์ ์ธ ์„ ์ˆœํ™˜ ํšจ๊ณผ(flywheel effect)๋ฅผ ์ฐฝ์ถœํ•  ๊ฒƒ์œผ๋กœ ์˜ˆ์ƒํ•ฉ๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ์‹ ์†ํ•œ ๋Œ€์‘ ์™ธ์—๋„ ์ฐจ๋ณ„ํ™”๋œ ๋ชจ๋ธ์„ ๊ฐœ๋ฐœํ•  ์ˆ˜ ์žˆ์—ˆ์œผ๋ฉฐ, ์ด๋Š” ์ง€๋‚œ ๋ช‡ ๋‹ฌ๊ฐ„ ๋ฐ˜๋ณต์ ์œผ๋กœ ๊ฒ€์ฆ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋ง์”€๋“œ๋ฆฐ ๋ฐ”์™€ ๊ฐ™์ด, ์‹œ์žฅ์€ ๋งค์šฐ ํฌ๊ณ  ๊ธฐ์ˆ  ๊ฒฝ๋กœ๋Š” ๋‹ค์–‘ํ•˜๊ฒŒ ๋ถ„ํ™”๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์ €ํฌ์—๊ฒŒ ์ค‘์š”ํ•œ ๊ฒƒ์€ ๊ธฐ์ˆ  ๋กœ๋“œ๋งต(technical roadmap)์„ ์ •์˜ํ•  ์ˆ˜ ์žˆ๋Š” ๋Šฅ๋ ฅ์ด ์žˆ๋Š”์ง€ ์—ฌ๋ถ€์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ๋ชจ๋“  ์˜์—ญ์—์„œ ์Šน๋ฆฌํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋Œ€์‹ , ์ €ํฌ์˜ ๋…์ ์ ์ธ ๊ฐ•์ (distinct strength)์„ ๋ณด์—ฌ์ค„ ์ˆ˜ ์žˆ๋Š” ๋ชจ๋ธ ์—ญ๋Ÿ‰(model capabilities)์„ ์ •์˜ํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. M2, Hailuo 2, Speech 2 ์‹œ๋ฆฌ์ฆˆ ๋ชจ๋ธ๋“ค์€ ๊ฐ๊ฐ ๋ช…ํ™•ํ•œ ์ฐจ๋ณ„์ (differentiation)์„ ๊ตฌ์ถ•ํ•˜์—ฌ ์‹œ์žฅ์—์„œ ๋น ๋ฅด๊ฒŒ ์ž…์ง€๋ฅผ ํ™•๋ณดํ•  ์ˆ˜ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ ํŠน์ง•์€ ๋‚ฎ์€ ์ง€์—ฐ ์‹œ๊ฐ„(latency)๊ณผ ๋†’์€ ๋น„์šฉ ํšจ์œจ์„ฑ(cost efficiency)์ž…๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํŠน์ง•๋“ค์ด ์ €ํฌ๋ฅผ ์ฐจ๋ณ„ํ™”์‹œํ‚ค๊ณ  ๋” ํฐ ์‹œ์žฅ ์ ์œ ์œจ(market share)์„ ํ™•๋ณดํ•˜๋Š” ๋ฐ ๊ธฐ์—ฌํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์ €ํฌ ์กฐ์ง๊ณผ ์ž์›์ด ๊ณ„์†ํ•ด์„œ ํ™•์žฅ๋จ์— ๋”ฐ๋ผ, ๋ชจ๋ธ ์ง„ํ™”์™€ ๊ธฐ์ˆ  ๋กœ๋“œ๋งต(technical roadmap)์— ๋Œ€ํ•œ ์ €ํฌ์˜ ๊นŠ์€ ์ดํ•ด๋Š” ์ด๋Ÿฌํ•œ ์ฐจ๋ณ„์ ๊ณผ ๊ทธ ๊ฐ€์น˜๋ฅผ ๋”์šฑ ๊ฐ•ํ™”ํ•  ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์š”์•ฝํ•˜์ž๋ฉด, ์ €ํฌ๋Š” ์‹œ์žฅ ์ ์œ ์œจ์„ ๋”์šฑ ๋Š˜๋ฆฌ๊ณ  ์—์ด์ „ํŠธ(agent)์™€ ๋” ๋„“์€ ์ƒ์‚ฐ์„ฑ ์‹œ์žฅ(productivity market)์„ ํ†ตํ•ด ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ ๋” ๋งŽ์€ ๋ŒํŒŒ๊ตฌ(breakthrough)๋ฅผ ๋งˆ๋ จํ•  ๊ฒƒ์ด๋ผ๊ณ  ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ €ํฌ๋Š” ๋” ๋น ๋ฅธ ๋ฐ˜๋ณต(iteration)๊ณผ ๋”์šฑ ๊ฐ•๋ ฅํ•œ ์ฐจ๋ณ„ํ™”๋œ ์œ„์น˜๋ฅผ ํ†ตํ•ด ๋” ํฐ ์‹œ์žฅ ์ ์œ ์œจ์„ ํ™•๋ณดํ•˜๋Š” ๋ฐ ์žˆ์–ด ๋” ํฐ ๋ŒํŒŒ๊ตฌ๋ฅผ ๋งŒ๋“ค๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค.

๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.
Operator: Your next question comes from Goldman Sachs.**Operator:** ๋‹ค์Œ ์งˆ๋ฌธ์€ ๊ณจ๋“œ๋งŒ์‚ญ์Šค์ž…๋‹ˆ๋‹ค.
Unknown Analyst: Thank you for your sharing. We know in this industry, there are tech giants, start-ups and open-source models. I would like to know where do you compete? What are your priorities?**Unknown Analyst:** ์„ค๋ช… ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ด ์—…๊ณ„์—๋Š” ๊ฑฐ๋Œ€ ๊ธฐ์ˆ  ๊ธฐ์—…(tech giants), ์Šคํƒ€ํŠธ์—…, ๊ทธ๋ฆฌ๊ณ  ์˜คํ”ˆ์†Œ์Šค ๋ชจ๋ธ๋“ค์ด ๋‹ค์–‘ํ•˜๊ฒŒ ์กด์žฌํ•œ๋‹ค๋Š” ๊ฒƒ์„ ์ž˜ ์•Œ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ™˜๊ฒฝ ์†์—์„œ ๊ท€์‚ฌ๋Š” ์–ด๋–ค ๋ถ„์•ผ์—์„œ ๊ฒฝ์Ÿํ•˜๊ณ  ๊ณ„์‹ ์ง€, ๊ทธ๋ฆฌ๊ณ  ํ•ต์‹ฌ์ ์ธ ์šฐ์„ ์ˆœ์œ„๋Š” ๋ฌด์—‡์ธ์ง€ ๋ง์”€ํ•ด์ฃผ์‹œ๋ฉด ๊ฐ์‚ฌํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค.
Junjie Yan: Founder, Executive Chairman, CTO & CEO As mentioned earlier, we are building and hoping to become an AI era platform company, driven by the continuous increase in intelligent density combined with a scalable, commercial growth. Compared to other AI companies, we differentiate in several ways. First, our strategic positioning. From day 1, we have focused on full-modality models to increase intelligence density and expand boundaries, creating differentiated value. At the same time, we build scalable products and business around model intelligent density, concentrating our resources on areas where we create different value. For example, in 2023, we decided not to build a general mobile assistant, i.e., Doubao and ChatGPT. We decided not to build a product like that because we did not believe we would create a distinctive value in this space. Instead, we focus on differentiated model R&D and product innovation rather than burning cash. Take our Hailuo and our MiniMax Agent products. These are our focuses. So this strategic decision reinforces our differentiation and increases our win rate. Another example is our commitment from day 1 to developing foundation models across the 3 models. As mentioned earlier, accumulation in each model is crucial. We have now reached a key stage of cross-modality integration. This position us advantageously in the inevitable trend towards full-modality fusion. And secondly, I want to talk about our R&D efficiency. In the AI era, success is not ultimately determined by how much money or resources you burn, but the speed at which intelligence improves. And that speed comes from R&D efficiency because that will turn into bigger market share and bigger -- higher efficiency. So we have been emphasizing on that and executing on that. We apply that across every stage of R&D, including algorithm optimization, experiment design, iteration cycles and et cetera. We fully leverage our agile organization structure, combining top-down and bottom-up approaches while using experience and infrastructure across modalities. So that ensures that we are always maintaining our leadership. In the long run, we believe that globally, only a small number of AI platform companies will automatically leave the industry. And we are one of the few independent companies with both meaningful advantages and a clear differentiation to win.**Junjie Yan:** ์•ž์„œ ๋ง์”€๋“œ๋ฆฐ ๋ฐ”์™€ ๊ฐ™์ด, ์ €ํฌ๋Š” ์ง€๋Šฅ ๋ฐ€๋„(intelligent density)์˜ ์ง€์†์ ์ธ ์ฆ๊ฐ€์™€ ํ™•์žฅ ๊ฐ€๋Šฅํ•œ(scalable) ์ƒ์—…์  ์„ฑ์žฅ(commercial growth)์„ ๊ฒฐํ•ฉํ•˜์—ฌ AI ์‹œ๋Œ€ ํ”Œ๋žซํผ ๊ธฐ์—…(AI era platform company)์ด ๋˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ๋‹ค๋ฅธ AI ๊ธฐ์—…๋“ค๊ณผ ๋น„๊ตํ–ˆ์„ ๋•Œ, ์ €ํฌ๋Š” ์—ฌ๋Ÿฌ ๋ฉด์—์„œ ์ฐจ๋ณ„์ ์„ ๊ฐ€์ง‘๋‹ˆ๋‹ค.

์ฒซ์งธ, ์ €ํฌ์˜ ์ „๋žต์  ํฌ์ง€์…”๋‹(strategic positioning)์ž…๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ฒซ๋‚ ๋ถ€ํ„ฐ ์ง€๋Šฅ ๋ฐ€๋„๋ฅผ ๋†’์ด๊ณ  ๊ฒฝ๊ณ„๋ฅผ ํ™•์žฅํ•˜๊ธฐ ์œ„ํ•ด ํ’€ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ๋ชจ๋ธ(full-modality models)์— ์ง‘์ค‘ํ•˜์—ฌ ์ฐจ๋ณ„ํ™”๋œ ๊ฐ€์น˜๋ฅผ ์ฐฝ์ถœํ•ด์™”์Šต๋‹ˆ๋‹ค. ๋™์‹œ์—, ์ €ํฌ๋Š” ๋ชจ๋ธ ์ง€๋Šฅ ๋ฐ€๋„๋ฅผ ์ค‘์‹ฌ์œผ๋กœ ํ™•์žฅ ๊ฐ€๋Šฅํ•œ ์ œํ’ˆ๊ณผ ๋น„์ฆˆ๋‹ˆ์Šค๋ฅผ ๊ตฌ์ถ•ํ•˜๋ฉฐ, ์ €ํฌ๊ฐ€ ์ฐจ๋ณ„ํ™”๋œ ๊ฐ€์น˜๋ฅผ ์ฐฝ์ถœํ•˜๋Š” ์˜์—ญ์— ์ž์›์„ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์˜ˆ๋ฅผ ๋“ค์–ด, 2023๋…„์—๋Š” Doubao๋‚˜ ChatGPT์™€ ๊ฐ™์€ ์ผ๋ฐ˜์ ์ธ ๋ชจ๋ฐ”์ผ ์–ด์‹œ์Šคํ„ดํŠธ(mobile assistant)๋ฅผ ๊ตฌ์ถ•ํ•˜์ง€ ์•Š๊ธฐ๋กœ ๊ฒฐ์ •ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ๊ทธ ๋ถ„์•ผ์—์„œ ์ฐจ๋ณ„ํ™”๋œ ๊ฐ€์น˜๋ฅผ ์ฐฝ์ถœํ•  ์ˆ˜ ์—†์„ ๊ฒƒ์ด๋ผ๊ณ  ํŒ๋‹จํ–ˆ๊ธฐ ๋•Œ๋ฌธ์— ๊ทธ๋Ÿฐ ์ œํ’ˆ์„ ๋งŒ๋“ค์ง€ ์•Š๊ธฐ๋กœ ๊ฒฐ์ •ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋Œ€์‹ ์—, ํ˜„๊ธˆ์„ ์†Œ์ง„ํ•˜๊ธฐ๋ณด๋‹ค๋Š” ์ฐจ๋ณ„ํ™”๋œ ๋ชจ๋ธ ์—ฐ๊ตฌ ๊ฐœ๋ฐœ(R&D)๊ณผ ์ œํ’ˆ ํ˜์‹ ์— ์ง‘์ค‘ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ์˜ ํ•˜์ด๋ฃจ์˜ค(Hailuo)์™€ ๋ฏธ๋‹ˆ๋งฅ์Šค ์—์ด์ „ํŠธ(MiniMax Agent) ์ œํ’ˆ์„ ์˜ˆ๋กœ ๋“ค ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๊ฒƒ๋“ค์ด ์ €ํฌ์˜ ํ•ต์‹ฌ ์ง‘์ค‘ ๋ถ„์•ผ์ž…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด๋Ÿฌํ•œ ์ „๋žต์  ๊ฒฐ์ •์€ ์ €ํฌ์˜ ์ฐจ๋ณ„์„ฑ์„ ๊ฐ•ํ™”ํ•˜๊ณ  ์Šน๋ฅ ์„ ๋†’์—ฌ์ค๋‹ˆ๋‹ค.

๋˜ ๋‹ค๋ฅธ ์˜ˆ์‹œ๋Š” ์ €ํฌ๊ฐ€ ์ฒซ๋‚ ๋ถ€ํ„ฐ 3๊ฐ€์ง€ ๋ชจ๋ธ ์ „๋ฐ˜์— ๊ฑธ์ณ ๊ธฐ๋ฐ˜ ๋ชจ๋ธ(foundation model) ๊ฐœ๋ฐœ์— ์ „๋…ํ•ด์™”๋‹ค๋Š” ์ ์ž…๋‹ˆ๋‹ค. ์•ž์„œ ๋ง์”€๋“œ๋ ธ๋“ฏ์ด, ๊ฐ ๋ชจ๋ธ์—์„œ์˜ ์ถ•์ ์€ ๋งค์šฐ ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ด์ œ ํ•ต์‹ฌ ๋‹จ๊ณ„์— ๋„๋‹ฌํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ฐ”๋กœ ๊ต์ฐจ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ํ†ตํ•ฉ(cross-modality integration)์ž…๋‹ˆ๋‹ค. ์ด๋Š” ํ”ผํ•  ์ˆ˜ ์—†๋Š” ์ „์ฒด ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ ์œตํ•ฉ(full-modality fusion) ํŠธ๋ Œ๋“œ ์†์—์„œ ์ €ํฌ๋ฅผ ์œ ๋ฆฌํ•œ ์œ„์น˜์— ๋†“์ด๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ๋‘ ๋ฒˆ์งธ๋กœ, ์ €ํฌ์˜ ์—ฐ๊ตฌ ๊ฐœ๋ฐœ ํšจ์œจ์„ฑ์— ๋Œ€ํ•ด ๋ง์”€๋“œ๋ฆฌ๊ณ  ์‹ถ์Šต๋‹ˆ๋‹ค. AI ์‹œ๋Œ€์— ๊ถ๊ทน์ ์ธ ์„ฑ๊ณต์€ ์–ผ๋งˆ๋‚˜ ๋งŽ์€ ๋ˆ์ด๋‚˜ ์ž์›์„ ์†Œ๋ชจํ•˜๋А๋ƒ๊ฐ€ ์•„๋‹ˆ๋ผ, ์ง€๋Šฅ์ด ์–ผ๋งˆ๋‚˜ ๋น ๋ฅด๊ฒŒ ํ–ฅ์ƒ๋˜๋А๋ƒ์— ๋‹ฌ๋ ค์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ๊ทธ ์†๋„๋Š” R&D ํšจ์œจ์„ฑ(R&D efficiency)์—์„œ ๋น„๋กฏ๋ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๊ฒฐ๊ตญ ๋” ํฐ ์‹œ์žฅ ์ ์œ ์œจ(market share)๊ณผ ๋” ๋†’์€ ํšจ์œจ์„ฑ์œผ๋กœ ์ด์–ด์งˆ ๊ฒƒ์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ ์ €ํฌ๋Š” ์ด๋ฅผ ๊ณ„์† ๊ฐ•์กฐํ•ด์™”๊ณ , ์‹ค์ œ๋กœ ์‹คํ–‰์— ์˜ฎ๊ธฐ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ์ด๋ฅผ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์ตœ์ ํ™”(algorithm optimization), ์‹คํ—˜ ์„ค๊ณ„(experiment design), ๋ฐ˜๋ณต ์ฃผ๊ธฐ(iteration cycles) ๋“ฑ์„ ํฌํ•จํ•œ R&D์˜ ๋ชจ๋“  ๋‹จ๊ณ„์— ์ ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์• ์ž์ผ ์กฐ์ง ๊ตฌ์กฐ(agile organization structure)๋ฅผ ์ตœ๋Œ€ํ•œ ํ™œ์šฉํ•˜์—ฌ, ํ•˜ํ–ฅ์‹(top-down) ๋ฐ ์ƒํ–ฅ์‹(bottom-up) ์ ‘๊ทผ ๋ฐฉ์‹์„ ๊ฒฐํ•ฉํ•˜๊ณ  ๋‹ค์–‘ํ•œ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(modality) ์ „๋ฐ˜์˜ ๊ฒฝํ—˜๊ณผ ์ธํ”„๋ผ๋ฅผ ํ™œ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ €ํฌ๊ฐ€ ํ•ญ์ƒ ๋ฆฌ๋”์‹ญ์„ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ๋„๋ก ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค.

์žฅ๊ธฐ์ ์œผ๋กœ ๋ณผ ๋•Œ, ์ €ํฌ๋Š” ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ์†Œ์ˆ˜์˜ AI ํ”Œ๋žซํผ ๊ธฐ์—…๋งŒ์ด ์—…๊ณ„๋ฅผ ์„ ๋„ํ•˜๊ฒŒ ๋  ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๋Š” ๊ฒฝ์Ÿ์—์„œ ์Šน๋ฆฌํ•  ์ˆ˜ ์žˆ๋Š” ์ƒ๋‹นํ•œ ์ด์ (meaningful advantages)๊ณผ ๋ช…ํ™•ํ•œ ์ฐจ๋ณ„์ (clear differentiation)์„ ๋ชจ๋‘ ๊ฐ–์ถ˜ ๋ช‡ ์•ˆ ๋˜๋Š” ๋…๋ฆฝ์ ์ธ ํšŒ์‚ฌ ์ค‘ ํ•˜๋‚˜์ž…๋‹ˆ๋‹ค.
Operator: Your next question from [ CIC ], [indiscernible].**Operator:** ๋‹ค์Œ ์งˆ๋ฌธ์€ CIC์—์„œ ์˜ค์‹ , ์„ฑํ•จ์ด ๋ถˆ๋ถ„๋ช…ํ•œ ๋ถ„๊ป˜ ๋ฐ›๊ฒ ์Šต๋‹ˆ๋‹ค.
Unknown Analyst: Congratulations on the strong results. You mentioned that in the first 2 months of 2026, token consumption for the M2 series is already 6x that of December last year. Is this explosive growth a onetime dividend or the beginning of a sustainable long-term trend? Because as we have noticed the explosion in the token consumption on OpenClaw. So that's why I'm asking this question. Do you think this is a onetime phenomenon or beginning of a long-term trend?**Unknown Analyst:** ์‹ค์  ์ถ•ํ•˜๋“œ๋ฆฝ๋‹ˆ๋‹ค. 2026๋…„ ์ฒซ ๋‘ ๋‹ฌ ๋™์•ˆ M2 ์‹œ๋ฆฌ์ฆˆ์˜ ํ† ํฐ ์†Œ๋น„๋Ÿ‰(token consumption)์ด ์ด๋ฏธ ์ž‘๋…„ 12์›” ๋Œ€๋น„ 6๋ฐฐ์— ๋‹ฌํ•œ๋‹ค๊ณ  ์–ธ๊ธ‰ํ•˜์…จ์Šต๋‹ˆ๋‹ค. ์ €ํฌ๊ฐ€ OpenClaw์˜ ํ† ํฐ ์†Œ๋น„๋Ÿ‰ ํญ์ฆ์„ ์ด๋ฏธ ํ™•์ธํ–ˆ๊ธฐ ๋•Œ๋ฌธ์— ๋“œ๋ฆฌ๋Š” ์งˆ๋ฌธ์ธ๋ฐ์š”. ์ด๋Ÿฌํ•œ ํญ๋ฐœ์ ์ธ ์„ฑ์žฅ์ด ์ผํšŒ์„ฑ ์ด์ต(onetime dividend)์ธ์ง€, ์•„๋‹ˆ๋ฉด ์ง€์† ๊ฐ€๋Šฅํ•œ ์žฅ๊ธฐ์  ์ถ”์„ธ์˜ ์‹œ์ž‘์ด๋ผ๊ณ  ๋ณด์‹œ๋Š”์ง€ ๊ถ๊ธˆํ•ฉ๋‹ˆ๋‹ค. ์ด๊ฒƒ์ด ์ผํšŒ์„ฑ ํ˜„์ƒ(onetime phenomenon)์ด๋ผ๊ณ  ๋ณด์‹œ๋Š”์ง€, ์•„๋‹ˆ๋ฉด ์žฅ๊ธฐ์ ์ธ ์ถ”์„ธ์˜ ์‹œ์ž‘์ด๋ผ๊ณ  ๋ณด์‹œ๋Š”์ง€์š”?
Junjie Yan: Founder, Executive Chairman, CTO & CEO Thank you for your question. We see this as the beginning of a long-term trend rather than a onetime dividend. But of course, the growth in the industry tends to follow a step function pattern where it's not moving linearly. We are able to constantly launching new models enable us to capture industry opportunities. I think a core part is our R&D strategy, is preparing resources capabilities in advance and defining every generation of model based on our understanding of how intelligence evolves. So other than the M2 model, the next wave of growth will come from several -- is underpinned by several factors. Actually, starting from second half of 2025, we have been proactively preparing for capabilities that will capture the multiple high-impact product market opportunities emerging in 2026. We believe the growth will be increasingly diversified. Coding still has a significant headroom. Though I mean, it's quite a decent tool as an assistant, we believe that it will continue to improve and will evolve from a system-level tool toward a colleague-level collaborator, even toward higher order intelligence, intelligent operator. So from our tech reserve and also R&D progress and our judgment, we believe what we mentioned will likely happen this year. And the second point is on workspace scenarios because this is a far larger end market and far broader market compared to coding. And there are a lot of more complicated problems or issues that involves many professions, use of a wide array of tools. It also -- many of the tasks conducted in these professions cannot be verified, and these created challenges, and we have been proactively preparing for such challenges. We expected that in workplace, we are going to rapid pace of progress we have seen in coding. And moving on to the multimodal domain. We believe that we are going to significantly lower the barrier to adoption in producing better models that would produce production-ready and longer videos. So model competition involves wins and losses with every company faces this reality. No company can guarantee permanent state-of-the-art leadership. However, we're confident in our ability to continue winning in these areas that matter most. I think a key strategy for us is to pushing the technical boundaries and leveraging that breakthrough to create a bigger ecosystem underpinned by our products and models. And as a final goal, leveraging that to capture the dividends. And we are confident in growing along with this industry, scaling our model differentiation, R&D efficiency, product innovation capabilities and global monetization capabilities into enduring organizational competitive advantages.**Junjie Yan:** ์งˆ๋ฌธ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ด๊ฒƒ์„ ์ผํšŒ์„ฑ ๋ฐฐ๋‹น(onetime dividend)์ด๋ผ๊ธฐ๋ณด๋‹ค๋Š” ์žฅ๊ธฐ์ ์ธ ์ถ”์„ธ(long-term trend)์˜ ์‹œ์ž‘์œผ๋กœ ๋ณด๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฌผ๋ก , ์‚ฐ์—…์˜ ์„ฑ์žฅ์€ ์„ ํ˜•์ ์œผ๋กœ ์›€์ง์ด๊ธฐ๋ณด๋‹ค๋Š” ๊ณ„๋‹จ ํ•จ์ˆ˜(step function) ํŒจํ„ด์„ ๋”ฐ๋ฅด๋Š” ๊ฒฝํ–ฅ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ์ €ํฌ๋Š” ์ง€์†์ ์œผ๋กœ ์‹ ๊ทœ ๋ชจ๋ธ์„ ์ถœ์‹œํ•จ์œผ๋กœ์จ ์‚ฐ์—… ๊ธฐํšŒ๋ฅผ ํฌ์ฐฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

ํ•ต์‹ฌ์ ์ธ ๋ถ€๋ถ„์€ ์ €ํฌ์˜ R&D ์ „๋žต(R&D strategy)์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์ž์›๊ณผ ์—ญ๋Ÿ‰์„ ๋ฏธ๋ฆฌ ์ค€๋น„ํ•˜๊ณ , ์ง€๋Šฅ์ด ์–ด๋–ป๊ฒŒ ์ง„ํ™”ํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ์ €ํฌ์˜ ์ดํ•ด๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ๊ฐ ์„ธ๋Œ€์˜ ๋ชจ๋ธ์„ ์ •์˜ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ M2 ๋ชจ๋ธ ์™ธ์—๋„, ๋‹ค์Œ ์„ฑ์žฅ ๋ฌผ๊ฒฐ(next wave of growth)์€ ์—ฌ๋Ÿฌ ์š”์ธ์— ์˜ํ•ด ๋’ท๋ฐ›์นจ๋  ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์‚ฌ์‹ค, 2025๋…„ ํ•˜๋ฐ˜๊ธฐ๋ถ€ํ„ฐ 2026๋…„์— ๋‚˜ํƒ€๋‚  ๋‹ค์ˆ˜์˜ ์˜ํ–ฅ๋ ฅ ์žˆ๋Š” ์ œํ’ˆ ์‹œ์žฅ ๊ธฐํšŒ(product market opportunities)๋ฅผ ํฌ์ฐฉํ•˜๊ธฐ ์œ„ํ•œ ์—ญ๋Ÿ‰์„ ์„ ์ œ์ ์œผ๋กœ ์ค€๋น„ํ•ด์™”์Šต๋‹ˆ๋‹ค. ์„ฑ์žฅ์€ ์ ์ฐจ ๋‹ค๊ฐํ™”๋  ๊ฒƒ์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.

์ฝ”๋”ฉ์€ ์—ฌ์ „ํžˆ ์ƒ๋‹นํ•œ ์„ฑ์žฅ ์—ฌ๋ ฅ(headroom)์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋ฌผ๋ก  ๋ณด์กฐ ๋„๊ตฌ(assistant tool)๋กœ์„œ๋„ ๊ฝค ๊ดœ์ฐฎ์€ ์ˆ˜์ค€์ด์ง€๋งŒ, ์ €ํฌ๋Š” ์ด๊ฒƒ์ด ๊ณ„์†ํ•ด์„œ ๋ฐœ์ „ํ•˜์—ฌ ์‹œ์Šคํ…œ ์ˆ˜์ค€์˜ ๋„๊ตฌ์—์„œ ๋™๋ฃŒ ์ˆ˜์ค€์˜ ํ˜‘๋ ฅ์ž(collaborator)๋กœ, ๋‚˜์•„๊ฐ€ ๋” ๊ณ ์ฐจ์›์ ์ธ ์ง€๋Šฅ(higher order intelligence)์„ ๊ฐ€์ง„ ์ง€๋Šฅํ˜• ์šด์˜์ž(intelligent operator)๋กœ ์ง„ํ™”ํ•  ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ €ํฌ์˜ ๊ธฐ์ˆ  ๋น„์ถ•(tech reserve)๊ณผ R&D ์ง„ํ–‰ ์ƒํ™ฉ, ๊ทธ๋ฆฌ๊ณ  ํŒ๋‹จ์— ๋น„์ถ”์–ด ๋ณผ ๋•Œ, ์ €ํฌ๊ฐ€ ๋ง์”€๋“œ๋ฆฐ ๋ฐ”๊ฐ€ ์˜ฌํ•ด ์•ˆ์— ์‹คํ˜„๋  ๊ฐ€๋Šฅ์„ฑ์ด ๋†’๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ๋‘ ๋ฒˆ์งธ ์š”์ ์€ ์ž‘์—… ๊ณต๊ฐ„ ์‹œ๋‚˜๋ฆฌ์˜ค(workspace scenarios)์— ๊ด€ํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ์ฝ”๋”ฉ ์‹œ์žฅ์— ๋น„ํ•ด ํ›จ์”ฌ ๋” ํฐ ์ตœ์ข… ์‹œ์žฅ(end market)์ด์ž ํ›จ์”ฌ ๋” ๊ด‘๋ฒ”์œ„ํ•œ ์‹œ์žฅ์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์—ฌ๋Ÿฌ ์ „๋ฌธ ๋ถ„์•ผ๊ฐ€ ๊ด€๋ จ๋˜์–ด ์žˆ๊ณ  ๋‹ค์–‘ํ•œ ๋„๊ตฌ๋“ค์„ ํ™œ์šฉํ•ด์•ผ ํ•˜๋Š” ๋” ๋ณต์žกํ•œ ๋ฌธ์ œ๋“ค์ด ๋งŽ์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ด๋Ÿฌํ•œ ์ „๋ฌธ ๋ถ„์•ผ์—์„œ ์ˆ˜ํ–‰๋˜๋Š” ๋งŽ์€ ์ž‘์—…๋“ค์€ ๊ฒ€์ฆํ•˜๊ธฐ ์–ด๋ ต๋‹ค๋Š” ์ ์ด ๋„์ „ ๊ณผ์ œ๋กœ ์ž‘์šฉํ–ˆ์œผ๋ฉฐ, ์ €ํฌ๋Š” ์ด๋Ÿฌํ•œ ๋„์ „ ๊ณผ์ œ๋“ค์— ๋Œ€ํ•ด ์„ ์ œ์ ์œผ๋กœ ์ค€๋น„ํ•ด ์™”์Šต๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ์—…๋ฌด ํ™˜๊ฒฝ์—์„œ๋„ ์ฝ”๋”ฉ ๋ถ„์•ผ์—์„œ ๋ณด์•˜๋˜ ๊ฒƒ๊ณผ ๊ฐ™์€ ๋น ๋ฅธ ์†๋„์˜ ๋ฐœ์ „์ด ์ด๋ฃจ์–ด์งˆ ๊ฒƒ์ด๋ผ๊ณ  ์˜ˆ์ƒํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ(multimodal) ์˜์—ญ์œผ๋กœ ํ™•์žฅํ•˜๋ฉด์„œ, ์ƒ์—…์ ์œผ๋กœ ํ™œ์šฉ ๊ฐ€๋Šฅํ•œ(production-ready) ๋” ๊ธด ๋น„๋””์˜ค๋ฅผ ์ œ์ž‘ํ•˜๋Š” ๋” ๋‚˜์€ ๋ชจ๋ธ์„ ๋งŒ๋“ค๋ฉด์„œ ๋„์ž… ์žฅ๋ฒฝ์„ ํฌ๊ฒŒ ๋‚ฎ์ถœ ์ˆ˜ ์žˆ์„ ๊ฒƒ์ด๋ผ๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค.

๋”ฐ๋ผ์„œ ๋ชจ๋ธ ๊ฒฝ์Ÿ์—๋Š” ์ŠนํŒจ๊ฐ€ ๋”ฐ๋ฅด๋ฉฐ, ๋ชจ๋“  ๊ธฐ์—…์ด ์ด๋Ÿฌํ•œ ํ˜„์‹ค์— ์ง๋ฉดํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์–ด๋–ค ๊ธฐ์—…๋„ ์˜๊ตฌ์ ์ธ ์ตœ์ฒจ๋‹จ(state-of-the-art) ์„ ๋‘ ์ž๋ฆฌ๋ฅผ ๋ณด์žฅํ•  ์ˆ˜๋Š” ์—†์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ €ํฌ๋Š” ๊ฐ€์žฅ ์ค‘์š”ํ•œ ํ•ต์‹ฌ ๋ถ„์•ผ์—์„œ ๊ณ„์†ํ•ด์„œ ์šฐ์œ„๋ฅผ ์ ํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ์˜ ํ•ต์‹ฌ ์ „๋žต์€ ๊ธฐ์ˆ ์  ํ•œ๊ณ„๋ฅผ ๋›ฐ์–ด๋„˜๊ณ , ๊ทธ๋Ÿฌํ•œ ๊ธฐ์ˆ ์  ๋ŒํŒŒ๊ตฌ๋ฅผ ํ™œ์šฉํ•˜์—ฌ ์ €ํฌ์˜ ์ œํ’ˆ๊ณผ ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋Š” ๋” ํฐ ์ƒํƒœ๊ณ„(ecosystem)๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

๊ถ๊ทน์ ์œผ๋กœ๋Š” ์ด๋ฅผ ํ™œ์šฉํ•˜์—ฌ ๊ทธ ๊ฒฐ์‹ค์„ ํ™•๋ณดํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๋Š” ์ด ์‚ฐ์—…๊ณผ ํ•จ๊ป˜ ์„ฑ์žฅํ•˜๋ฉฐ, ์ €ํฌ์˜ ๋ชจ๋ธ ์ฐจ๋ณ„ํ™”(model differentiation), R&D ํšจ์œจ์„ฑ(R&D efficiency), ์ œํ’ˆ ํ˜์‹  ์—ญ๋Ÿ‰(product innovation capabilities), ๊ทธ๋ฆฌ๊ณ  ๊ธ€๋กœ๋ฒŒ ์ˆ˜์ตํ™” ์—ญ๋Ÿ‰(global monetization capabilities)์„ ์ง€์†์ ์ธ ์กฐ์ง์˜ ๊ฒฝ์Ÿ ์šฐ์œ„(competitive advantages)๋กœ ๋ฐœ์ „์‹œํ‚ฌ ์ˆ˜ ์žˆ๋‹ค๊ณ  ํ™•์‹ ํ•ฉ๋‹ˆ๋‹ค.
Operator: And your next question comes from Thomas Chong of Jefferies.**Operator:** ๋‹ค์Œ ์งˆ๋ฌธ์€ ์ œํ”„๋ฆฌ์Šค(Jefferies)์˜ ํ† ๋งˆ์Šค ์ด๋‹˜์ž…๋‹ˆ๋‹ค.
Thomas Chong: Jefferies LLC, Research Division You mentioned that internal agent interns now nearly covers 90% of employees. So what insight has this change brought to you? And how does it feed back into your product and technology development?**Thomas Chong:** ๋‚ด๋ถ€ ์—์ด์ „ํŠธ ์ธํ„ด์ด ์ด์ œ ๊ฑฐ์˜ ์ง์› 90%๋ฅผ ์ฐจ์ง€ํ•œ๋‹ค๊ณ  ๋ง์”€ํ•˜์…จ์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ณ€ํ™”๋ฅผ ํ†ตํ•ด ์–ด๋–ค ํ†ต์ฐฐ๋ ฅ์„ ์–ป์œผ์…จ๋Š”์ง€ ๊ถ๊ธˆํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ด ์ ์ด ๊ท€์‚ฌ์˜ ์ œํ’ˆ ๋ฐ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ์— ์–ด๋–ป๊ฒŒ ํ”ผ๋“œ๋ฐฑ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๊นŒ?
Junjie Yan: Founder, Executive Chairman, CTO & CEO Thank you for your question. So we are not only an AI company. We aim to build a truly AI-native platform company. While researching AI models, we want to turn ourselves into an AI-native company. That's a key goal for us as an organization. And there are 2 things we're focusing on. Number one, speed. That is the speed of progress, by that, I mean. I think the fundamental reason why we want to be AI-native is that we have limited resources as a start-up. We need to maximize our efficiency so as to survive and achieve success. So we have been leveraging AI agent in turn, and many of our employees are using that in their day-to-day work. And we have observed a clear trend. In many cases, the dynamics shifting from people teaching agents how to work, to people observing how agents work. And at times, agents even surprise us. So this has not only shortened our organizational workflows, but also allowed every stage to benefit from improvements in intelligence. From model iteration and product and innovation to customer service, our feedback and iteration looks are accelerating. And at the same time, our employees can focus more on higher-value work, further accelerating how we think and innovate as an organization. So this also feedbacks to our R&D of models because this allow us to define model-intelligence objectives. For example, when agents are deployed within the company, we can clearly observe that. Even the best models today still get things wrong or couldn't get things done properly. And these gaps exactly show highest economic value, and they inform the R&D priorities for the next generation of models and agents. This enable us to define our objectives more clearly. So the more we deploy these agents, I mean, the clear direction we have on the iteration of these models. Over the past few months, our model iteration speed, revenue growth, customer service capability and token throughput have all improved. This allow us to define new model objectives faster. We are maximizing the value of AI internally within the company. We believe that for goal, like we said, building an AI-native company, we are seeing positive flywheel already within the company. And I believe this will become one of the key competitive advantages of our organization.**Junjie Yan:** ์งˆ๋ฌธ ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ๋‹จ์ˆœํžˆ AI ๊ธฐ์—…์ด ์•„๋‹™๋‹ˆ๋‹ค. ์ง„์ •ํ•œ AI ๋„ค์ดํ‹ฐ๋ธŒ(AI-native) ํ”Œ๋žซํผ ๊ธฐ์—…์„ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. AI ๋ชจ๋ธ์„ ์—ฐ๊ตฌํ•˜๋ฉด์„œ, ์ €ํฌ ์Šค์Šค๋กœ๋ฅผ AI ๋„ค์ดํ‹ฐ๋ธŒ ๊ธฐ์—…์œผ๋กœ ์ „ํ™˜ํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ด๊ฒƒ์ด ์ €ํฌ ์กฐ์ง์˜ ํ•ต์‹ฌ ๋ชฉํ‘œ์ž…๋‹ˆ๋‹ค.

๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๊ฐ€ ์ง‘์ค‘ํ•˜๊ณ  ์žˆ๋Š” ๋‘ ๊ฐ€์ง€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ์ฒซ์งธ๋Š” ์†๋„์ž…๋‹ˆ๋‹ค. ์—ฌ๊ธฐ์„œ ๋งํ•˜๋Š” ์†๋„๋Š” ๋ฐ”๋กœ ๋ฐœ์ „์˜ ์†๋„๋ฅผ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. ์ €ํฌ๊ฐ€ AI ๋„ค์ดํ‹ฐ๋ธŒ๊ฐ€ ๋˜๊ณ ์ž ํ•˜๋Š” ๊ทผ๋ณธ์ ์ธ ์ด์œ ๋Š” ์Šคํƒ€ํŠธ์—…(startup)์œผ๋กœ์„œ ์ž์›์ด ์ œํ•œ์ ์ด๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์ƒ์กดํ•˜๊ณ  ์„ฑ๊ณตํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ํšจ์œจ์„ฑ์„ ๊ทน๋Œ€ํ™”ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ž˜์„œ ์ €ํฌ๋Š” AI ์—์ด์ „ํŠธ(AI agent)๋ฅผ ํ™œ์šฉํ•ด ์™”๊ณ , ๋งŽ์€ ์ง์›๋“ค์ด ์ผ์ƒ ์—…๋ฌด์—์„œ ์ด๋ฅผ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ €ํฌ๋Š” ๋ถ„๋ช…ํ•œ ์ถ”์„ธ๋ฅผ ํ™•์ธํ–ˆ์Šต๋‹ˆ๋‹ค. ๋งŽ์€ ๊ฒฝ์šฐ, ์‚ฌ๋žŒ๋“ค์ด ์—์ด์ „ํŠธ(agent)์—๊ฒŒ ์ผํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ๊ฐ€๋ฅด์น˜๋˜ ๊ฒƒ์—์„œ ์—์ด์ „ํŠธ๊ฐ€ ์ผํ•˜๋Š” ๋ฐฉ์‹์„ ๊ด€์ฐฐํ•˜๋Š” ๊ฒƒ์œผ๋กœ ์–‘์ƒ์ด ๋ฐ”๋€Œ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋•Œ๋กœ๋Š” ์—์ด์ „ํŠธ๊ฐ€ ์ €ํฌ๋ฅผ ๋†€๋ผ๊ฒŒ ํ•˜๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ์กฐ์ง์˜ ์›Œํฌํ”Œ๋กœ์šฐ(workflow)๋ฅผ ๋‹จ์ถ•์‹œ์ผฐ์„ ๋ฟ๋งŒ ์•„๋‹ˆ๋ผ, ๋ชจ๋“  ๋‹จ๊ณ„์—์„œ ์ง€๋Šฅ ๊ฐœ์„ ์˜ ์ด์ ์„ ์–ป์„ ์ˆ˜ ์žˆ๋„๋ก ํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ชจ๋ธ(model) ๋ฐ˜๋ณต(iteration)๊ณผ ์ œํ’ˆ ๋ฐ ํ˜์‹ ๋ถ€ํ„ฐ ๊ณ ๊ฐ ์„œ๋น„์Šค์— ์ด๋ฅด๊ธฐ๊นŒ์ง€, ์ €ํฌ์˜ ํ”ผ๋“œ๋ฐฑ(feedback) ๋ฐ ๋ฐ˜๋ณต(iteration) ์ฃผ๊ธฐ๊ฐ€ ๊ฐ€์†ํ™”๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋™์‹œ์— ์ €ํฌ ์ง์›๋“ค์€ ๋” ๋†’์€ ๊ฐ€์น˜์˜ ์—…๋ฌด์— ์ง‘์ค‘ํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์—ˆ์œผ๋ฉฐ, ์ด๋Š” ์กฐ์ง์œผ๋กœ์„œ ์ €ํฌ๊ฐ€ ์‚ฌ๊ณ ํ•˜๊ณ  ํ˜์‹ ํ•˜๋Š” ๋ฐฉ์‹์„ ๋”์šฑ ๊ฐ€์†ํ™”ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋˜ํ•œ ๋ชจ๋ธ(model) ์—ฐ๊ตฌ ๊ฐœ๋ฐœ(R&D)์—๋„ ๋‹ค์‹œ ๋ฐ˜์˜๋˜๋Š”๋ฐ, ์ด๋ฅผ ํ†ตํ•ด ๋ชจ๋ธ ์ง€๋Šฅ ๋ชฉํ‘œ๋ฅผ ๋ช…ํ™•ํžˆ ์ •์˜ํ•  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ์˜ˆ๋ฅผ ๋“ค์–ด, ์—์ด์ „ํŠธ๊ฐ€ ํšŒ์‚ฌ ๋‚ด๋ถ€์— ๋ฐฐํฌ๋  ๋•Œ ์ €ํฌ๋Š” ์ด๋ฅผ ๋ช…ํ™•ํ•˜๊ฒŒ ๊ด€์ฐฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์˜ค๋Š˜๋‚  ์ตœ๊ณ ์˜ ๋ชจ๋ธ๋“ค๋„ ์—ฌ์ „ํžˆ ์˜ค๋ฅ˜๋ฅผ ๋ฒ”ํ•˜๊ฑฐ๋‚˜ ์ œ๋Œ€๋กœ ์ผ์„ ์ฒ˜๋ฆฌํ•˜์ง€ ๋ชปํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ด๋Ÿฌํ•œ ๋ถ€์กฑํ•œ ๋ถ€๋ถ„๋“ค์ด ๋ฐ”๋กœ ๊ฐ€์žฅ ๋†’์€ ๊ฒฝ์ œ์  ๊ฐ€์น˜๋ฅผ ๋ณด์—ฌ์ฃผ๋ฉฐ, ์ฐจ์„ธ๋Œ€ ๋ชจ๋ธ๊ณผ ์—์ด์ „ํŠธ ๊ฐœ๋ฐœ์„ ์œ„ํ•œ R&D(์—ฐ๊ตฌ ๊ฐœ๋ฐœ) ์šฐ์„ ์ˆœ์œ„๋ฅผ ๊ฒฐ์ •ํ•˜๋Š” ๋ฐ ์ค‘์š”ํ•œ ์ •๋ณด๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ์šฐ๋ฆฌ๋Š” ๋ชฉํ‘œ๋ฅผ ๋”์šฑ ๋ช…ํ™•ํ•˜๊ฒŒ ์ •์˜ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋”ฐ๋ผ์„œ ์ด๋Ÿฌํ•œ ์—์ด์ „ํŠธ๋“ค์„ ๋” ๋งŽ์ด ๋ฐฐํฌํ• ์ˆ˜๋ก, ์ด ๋ชจ๋ธ๋“ค์˜ ๋ฐ˜๋ณต(iteration) ๋ฐฉํ–ฅ์ด ๋”์šฑ ๋ช…ํ™•ํ•ด์ง€๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ง€๋‚œ ๋ช‡ ๋‹ฌ๊ฐ„ ์ €ํฌ ๋ชจ๋ธ์˜ ๋ฐ˜๋ณต(iteration) ์†๋„, ๋งค์ถœ ์„ฑ์žฅ, ๊ณ ๊ฐ ์„œ๋น„์Šค ์—ญ๋Ÿ‰, ๊ทธ๋ฆฌ๊ณ  ํ† ํฐ ์ฒ˜๋ฆฌ๋Ÿ‰(token throughput)์ด ๋ชจ๋‘ ํ–ฅ์ƒ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์ €ํฌ๊ฐ€ ์ƒˆ๋กœ์šด ๋ชจ๋ธ ๋ชฉํ‘œ๋ฅผ ๋” ๋น ๋ฅด๊ฒŒ ์ •์˜ํ•  ์ˆ˜ ์žˆ๋„๋ก ํ•ด์ค๋‹ˆ๋‹ค.

์ €ํฌ๋Š” ํšŒ์‚ฌ ๋‚ด๋ถ€์ ์œผ๋กœ AI์˜ ๊ฐ€์น˜๋ฅผ ๊ทน๋Œ€ํ™”ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋ง์”€๋“œ๋ ธ๋“ฏ์ด AI ๋„ค์ดํ‹ฐ๋ธŒ(AI-native) ํšŒ์‚ฌ๋ฅผ ๊ตฌ์ถ•ํ•˜๋ ค๋Š” ๋ชฉํ‘œ๋ฅผ ์œ„ํ•ด, ์ €ํฌ๋Š” ์ด๋ฏธ ํšŒ์‚ฌ ๋‚ด์—์„œ ๊ธ์ •์ ์ธ ์„ ์ˆœํ™˜(flywheel)์ด ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ๋‹ค๊ณ  ์ƒ๊ฐํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋ฆฌ๊ณ  ์ด๊ฒƒ์ด ์ €ํฌ ์กฐ์ง์˜ ํ•ต์‹ฌ ๊ฒฝ์Ÿ ์šฐ์œ„ ์ค‘ ํ•˜๋‚˜๊ฐ€ ๋  ๊ฒƒ์ด๋ผ๊ณ  ๋ฏฟ์Šต๋‹ˆ๋‹ค.
Unknown Executive: Thank you once again for joining us today. If you have any further questions, please contact our IR team at any time. Thank you. [Statements in English on this transcript were spoken by an interpreter present on the live call.]**Unknown Executive:** ์˜ค๋Š˜ ๋‹ค์‹œ ํ•œ๋ฒˆ ์ €ํฌ ์ปจํผ๋Ÿฐ์Šค์ฝœ์— ์ฐธ์„ํ•ด์ฃผ์…”์„œ ์ง„์‹ฌ์œผ๋กœ ๊ฐ์‚ฌ๋“œ๋ฆฝ๋‹ˆ๋‹ค.

์ถ”๊ฐ€์ ์ธ ์งˆ๋ฌธ์ด ์žˆ์œผ์‹œ๋ฉด ์–ธ์ œ๋“ ์ง€ ์ €ํฌ IRํŒ€(Investor Relations team)์œผ๋กœ ์—ฐ๋ฝ ์ฃผ์‹ญ์‹œ์˜ค. ๊ฐ์‚ฌํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ“Œ ์š”์•ฝ

๋‹ค์Œ์€ ์ „๋ฌธ ํˆฌ์ž์ž๋ฅผ ์œ„ํ•œ ์š”์•ฝ์ž…๋‹ˆ๋‹ค.

* **์„ฑ์žฅ ๋ฐ ์ „๋ง:** 2026๋…„ ์ดˆ M2 ์‹œ๋ฆฌ์ฆˆ ํ† ํฐ ์†Œ๋น„๋Ÿ‰์ด 2025๋…„ 12์›” ๋Œ€๋น„ 6๋ฐฐ ์ฆ๊ฐ€ํ•˜๋Š” ํญ๋ฐœ์ ์ธ ์„ฑ์žฅ์„ ๊ธฐ๋กํ–ˆ์œผ๋ฉฐ, ์ด๋Š” M3