Excited to announce that our bridge funding round includes a planned investment from @nvidia , alongside existing investors. Now back to work - we have a frontier model to ship. 🫡
Excited to announce that Motif’s latest funding round includes a planned investment from @nvidia , alongside existing investors.
Now we go all in on Motif 4, and we’ll be back with a best-in-class frontier LLM. Thank you to our investors for the support. 🙏
It has been great to see the community come together to build with Motif 3 - so we are extending free access on @InfronAI by two weeks, through October 1st.
Thank you to everyone who tried it and shared feedback. If you haven’t tried it yet, we’d appreciate giving it a look! 🙇♂️
Motif-3 stays FREE on Infron through October 1.
No minimum balance. Same API.
Try Motif-3: infron.ai/models/motif/motif…
Give Motif-3 a task that takes more than one step. Let your agent inspect a repo, run tests, and work through a failure.
His original post (in Korean): lnkd.in/p/gANjvBev
![[Motif-code를 공개합니다. 여러분의 도움이 필요합니다] 지난 8월 말, Motif Technologies 가 독파모(독자파운데이션모델 사업) 2차 평가에서 탈락했습니다....](https://pbs.twimg.com/card_img/2120398095565414400/XGvP6ZVI?format=jpg&name=800x419)
52
🔗 Motif-3 on Infron: infron.ai/models/motif/motif…
🔗 Motif-3 Technical report: huggingface.co/Motif-Technol…
2
102
Motif 3 is now public and available through an API for the first time. We've been heads-down on research. Now that you can build with it, we'd love to hear where it shines and where we can improve. Your feedback will shape what we build next.
Free on @InfronAI through Sep 29.
Motif 3 scores 35.3 on tau-3 Banking, ahead of DeepSeek V4 Pro in @motif_tech's published comparison: infron.ai/models/motif/motif…
It activates just 13.2B parameters per token.
Free access on Infron. Report and API details in the first reply.
Junghwan Lim retweeted
I am told the Hodge Conjecture is very close to being verified by OpenAI, and that one of OpenAI or Anthropic are also close to solving Birch-Swinnerton-Dyer. The race to be 'next' behind the scenes is unlike anything I've had described to me before
If true - and it may not be, given the scale of the rumour mill right now - it could mean 3 Millennium Problems fall in the space of a month. Crazy times
Junghwan Lim retweeted
While everyone is raging at me for defending OpenAI/Anthropic in one narrow area, we have got loads of new, fun, and powerful open models. A far better place than your attention than frontier lab drama.
Artifacts 24: Motif-3, GLM-5.3, Hy4-preview and open model licenses. In this issue, the 25 models from july/aug/sept you should be aware of, from:
InclusionAI @TheInclusionAI (4)
Qwen @alibaba_qwen (3)
Z ai @Zai_org (2)
DeepSeek @deepseek_ai (2)
Motif Technologies @motif_tech
Dots Studio @dotsstudioai
Tencent @TencentGlobal
NVIDIA @NVIDIAAI
IBM @IBM
Institute of Foundation Models @IFM_AI
Trillion Labs @TrillionLabs
SenseTime @SenseTime_AI
Meta @AIatMeta
Liquid AI @liquidai
Cohere @Cohere_Labs
Superwhisper @superwhisper
Google @GoogleAI
MiniMax @MiniMax_AI
Read the issue below.
Junghwan Lim retweeted
Korea’s Trillion-Dollar Sovereign AI Investment: Nvidia Wins, Hynix Loses
Korea hosts a Squid Games, National AI Tournament,
the best non-Chinese open source model gets eliminated,
why Nvidia needs open source, implications for Hynix and Samsung
newsletter.semianalysis.com/…
Junghwan Lim retweeted
💥最新 AI 模型梯队排行榜08/31：中国模型全面紧追，第一梯队门口已挤满中国模型！
🔹夯：Opus 5 / GPT-5.6 SOL / Grok 4.6 / Kimi K3
🔹顶级：DeepSeek V4 Flash / GLM-5.3 / Qwen3.8 2.4T A95B / GLM-5.3-Flash / Hy-4 Preview / Qwen-3.8-Flash-Next
🔹人上人：GPT-5.6 Terra / Qwen3.8 27B / Motif3 / MiniMax-M3 / Ornith-1.5
🔹NPC：DeepSeek V4 Pro / GPT-5.6 Luna / mimo V2.5 Pro / Muse Spark 1.2
🔹拉完了：Gemini 3.7 Flash / Sonnet 5 / thinking / Nemotron 3 Ultra / Hy3 / Muse Glimmer
🔹落伍了：Gemini 3.6 Flash / Gemini 3.5 Flash / Haiku-4.5 / Mistral Medium 3.5 / Nemotron 3.5 Lightning
GLM-5.3-Flash、Hy-4 Preview、Qwen-3.8-Flash-Next 已贴身紧逼第一梯队，Kimi 直接站上夯位。Gemini 3.7/3.6/3.5 整建制掉到后面。格局变了。
#AI梯队 #中国大模型 #GLM53Flash #Hy4 #Qwen38 #KimiK3 #DeepSeek #Gemini掉队 #08月31日榜单 #从夯到拉
Junghwan Lim retweeted
📢 About Motif 3📢
▪️ Decoder-only MoE language model with 314B parameters, 13.2B active.
▪️ Scored 47 on @ArtificialAnlys Intelligence Index, 5th among all open-weight models, ahead of @MiniMax_AI M3, @thinkymachines Inkling, @nvidia Nemotron, and all other models developed outside of China.
▪️ Built from scratch around a novel attention architecture (GDLA) and trained on only 768 B200s in ~5 months.
▪️ MIT license.
(1/3)
Junghwan Lim retweeted
Jisoo Lee, one of the driving forces behind Motif 3, will share the story of its development journey and how NVIDIA technologies were leveraged along the way.
If you’re interested, register for the session below!
⏰ Time: Wednesday, August 26, 2:00–3:00 PM
📌 Register here: nvda.ws/4zFUgAK
Motif 3 개발의 주역이신 이지수님께서 Motif 3 개발 여정과 Nvidia 기술들을 어떻게 활용했는지에 대해서 설명해주실 예정입니다.
관심있으신 분들은 아래 링크에서 일정 등록하세요!
⏰ 일시 : 8월 26일(수) 오후 2시 - 3시
📌 일정 등록하기: nvda.ws/4zFUgAK
🔔 [D-1 LIVE] 내일 오후 2시! 내 도메인에 딱 맞는 AI 모델 구축 라이브 스트리밍
Nemotron 3.5 Lightning 기반 SFT·LoRA 파인튜닝부터 모델 평가, NeMo RL 미리보기까지 — NVIDIA 전문가가 실무 핵심을 단계별로 짚어드립니다. 스페셜 게스트 Motif Technologies의 Blackwell 기반 K-AI 오픈 모델 구축 비하인드도 놓치지 마세요.
⏰ 일시: 8월 26일(수) 오후 2시 - 3시
📌 일정 등록하기: nvda.ws/4gsfXuW"
Junghwan Lim retweeted
글로벌 가장 널리 인정받는 AAII가 한국을 명실공히 AI 레이스 No.3 국가로 인정하고 있습니다. 국가 AI 전략과 #혁신역량 보유 국내 #독파모 기업들이 함께 만든 성과입니다. 앞으로도 규모 무관 가장 혁신적이고 능력과 의지가 있는 기업에 집중 지원 투자해서 경쟁력을 높여가야 겠습니다.
South Korea has once again established a clear #3 position in the global AI race, trailing only the United States and China. The depth and vibrancy of Korea’s AI ecosystem is supported by strong domestic talent, public investment, and the incentives created by Korea’s Sovereign AI Foundation Model project
A number of South Korean labs, including Motif Technologies, Upstage, SK Telecom, and LG AI Research, have now scored above 30 on the Artificial Analysis Intelligence Index. This concentration of competitive model developers highlights the growing density of Korea’s domestic AI ecosystem. Among them, Motif Technologies and Upstage have models scoring above 40, making them the highest-scoring models on the Intelligence Index developed outside the United States and China.
Leading models from each lab include:
➤ Motif Technologies’ Motif 3 scored 47 on the Intelligence Index (314B total parameters, 13B active; open weights).
➤ Upstage’s Solar Pro 4 scored 42 on the Intelligence Index. It is a proprietary reasoning model.
➤ SK Telecom’s A.X K2 scored 35 on the Intelligence Index (692B total parameters, 33B active; open weights).
➤ LG AI Research’s K-EXAONE 2.0 scored 31 on the Intelligence Index (750B total parameters, 37B active; open weights).
Outside the competition, Korea’s model-development ecosystem has continued to broaden. Trillion Labs, Korea Telecom (KT), and other AI labs are building out their own model families, adding to a domestic field that is considerably deeper and more competitive than it was a year ago.
Junghwan Lim retweeted
South Korea has once again established a clear #3 position in the global AI race, trailing only the United States and China. The depth and vibrancy of Korea’s AI ecosystem is supported by strong domestic talent, public investment, and the incentives created by Korea’s Sovereign AI Foundation Model project
A number of South Korean labs, including Motif Technologies, Upstage, SK Telecom, and LG AI Research, have now scored above 30 on the Artificial Analysis Intelligence Index. This concentration of competitive model developers highlights the growing density of Korea’s domestic AI ecosystem. Among them, Motif Technologies and Upstage have models scoring above 40, making them the highest-scoring models on the Intelligence Index developed outside the United States and China.
Leading models from each lab include:
➤ Motif Technologies’ Motif 3 scored 47 on the Intelligence Index (314B total parameters, 13B active; open weights).
➤ Upstage’s Solar Pro 4 scored 42 on the Intelligence Index. It is a proprietary reasoning model.
➤ SK Telecom’s A.X K2 scored 35 on the Intelligence Index (692B total parameters, 33B active; open weights).
➤ LG AI Research’s K-EXAONE 2.0 scored 31 on the Intelligence Index (750B total parameters, 37B active; open weights).
Outside the competition, Korea’s model-development ecosystem has continued to broaden. Trillion Labs, Korea Telecom (KT), and other AI labs are building out their own model families, adding to a domestic field that is considerably deeper and more competitive than it was a year ago.
Junghwan Lim retweeted
Congratulations to the @motif_tech team on the release of Motif 3!
AfterQuery is proud to have served as the sole data partner on this model.
Junghwan Lim retweeted
in the last 24 hours
🇰🇷Korea finally can into AI w/ Motif 3
🇺🇸US: Grok 4.6, Gemini Flash 3.7
then suddenly🇨🇳China: Qwen3.8 Max, Qwen3.8 27B, GLM5.3, Deepseek v4 Pro 0813, dots3-note
Does China perhaps hold a bit of a grudge against SK? or Elon? cuz theyre lookin a bit salty rn😂
Junghwan Lim retweeted
Motif 3
A 314B-parameter MoE language model with only 13.2B active parameters per token, trained on 12.5T tokens with a novel Grouped Differential Latent Attention.
Junghwan Lim retweeted
Motif 3: Technical Report
Junghwan Lim, Joon Son Chung, Sungmin Lee, Wai Ting Cheung, Gihun Cho, Minsu Ha, Sangho Kang, Beomgyu Kim, Dongseok Kim, Jangwoong Kim, Taehyun Kim, Taewhan Kim, Jeesoo Lee, Jeongdoo Lee, Junhyeok Lee, Dongpin Oh, …
arxiv.org/abs/2608.09119 [𝚌𝚜.𝙰𝙸]
ALT We introduce Motif 3, a decoder-only Mixture-of-Experts language model with 314 billion total parameters and 13.2 billion activated per token. Each sparse MoE layer contains 384 routed experts, with eight selected per token. This fine-grained sparsity provides substantial expert capacity while limiting computation. Motif 3 is built around Grouped Differential Latent Attention (GDLA), which integrates grouped differential attention with the compressed key-value representation of Multi-head Latent Attention. The architecture further incorporates modified manifold-constrained hyper-connections, Expert Specific PolyNorm activations, and multi-token prediction to improve optimization stability, expert specialization, and inference efficiency. We pretrain Motif 3 on approximately 12.5 trillion tokens spanning web documents, STEM, code, mathematics, multilingual content, and domain-specialized corpora. Expert-balancing and numerical-stabilization techniques support stable training at scale, w
Junghwan Lim retweeted
Proud to have @motif_tech’s new model, Motif 3, on our ai& inference platform today.
@aiand_ are proud launch partners and excited about where sovereign AI is going.
Thanks to the Motif team, my co-founder @hara_shimpei, and all the team for the hard work to make this happen.
Motif 3 is officially here.
Today, we’re releasing Motif 3 Base and Motif 3.
Motif 3 Base is our pretrained foundation model, while Motif 3 is built on top of it with post-training using NVIDIA NeMo-RL.
Both models were developed with the support of Korea's Ministry of Science and trained on NVIDIA B200 GPUs.
With Motif 3, we’re taking our first step into the frontier LLM race.
There’s still a long way to go, but we believe we’ve built a model that deserves to be in the race.
We’ve shared the details in our technical report and blog post.
In collaboration with NVIDIA, we’re also releasing an NVFP4 version of Motif 3.
All models will be available on Hugging Face.
Links are in the comments.
Junghwan Lim retweeted
Congrats @motif_tech team! This release is a great example of Korea’s AI ecosystem moving fast.