Sounds like a major breech of National Security. Why is no one talking about this???
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
They are probably afraid of being embarrassed by their funny pronunciation. "Jipu"?!
do you have link of those estimates?
The time has arrived for pre-Chinese New Year model releases
First up is the new Qwen3-Max-Thinking model from Alibaba with benchmarks on par with the best US models.
Most notably, Qwen team seems to have discovered a new test-time scaling technique allowing it to significantly improve performance
First up is the new Qwen3-Max-Thinking model from Alibaba with benchmarks on par with the best US models.
Most notably, Qwen team seems to have discovered a new test-time scaling technique allowing it to significantly improve performance
well, it's not just a test, it's been officially released.
Released here:
Looks like the Chinese labs outside of Deep Seek are mostly moving to native multimodal models, following Western labs. Of course, Deep Seek v4 could also be a multimodal model, we'll just have to see.
Everybody is rushing to release models before Chinese New Years. Alibaba, Minimax, and now Moon Shot. z.AI released recently as well. That just leaves Deep Seek. If they manage to crash the Western stock market again before Chinese New Years, that'd result in some real fire works. But regardless, it's great to see Chinese models are at least keeping up on the bench marks. The real test, however, will be ecosystem, context length, and continual learning in 2026 - three important bottle necks to current foundation model performance.
Looks like the Chinese labs outside of Deep Seek are mostly moving to native multimodal models, following Western labs. Of course, Deep Seek v4 could also be a multimodal model, we'll just have to see.
Everybody is rushing to release models before Chinese New Years. Alibaba, Minimax, and now Moon Shot. z.AI released recently as well. That just leaves Deep Seek. If they manage to crash the Western stock market again before Chinese New Years, that'd result in some real fire works. But regardless, it's great to see Chinese models are at least keeping up on the bench marks. The real test, however, will be ecosystem, context length, and continual learning in 2026 - three important bottle necks to current foundation model performance.
Having tried the new Kimi K2.5 and Qwen Max-Thinking models I can only say... the gap between Qwen and Kimi is getting larger, not smaller. Kimi is simply in another league. Qwen not only did poorly, it straight-up hallucinated on several of my questions.
K2 was already a good model but it is now clearly great with the 2.5 update. The reasoning performance is on par with Opus 4.5 and GPT5.2-high in my testing.
Of all the major Chinese labs, Moonshot is probably most deserving of being flooded with GPUs and investment. Too bad geopolitics will prevent it. The talent density of the team is astounding.
K2 was already a good model but it is now clearly great with the 2.5 update. The reasoning performance is on par with Opus 4.5 and GPT5.2-high in my testing.
Of all the major Chinese labs, Moonshot is probably most deserving of being flooded with GPUs and investment. Too bad geopolitics will prevent it. The talent density of the team is astounding.
Agent Swarm is multi agent reinforcement learning right ?? Should be very compute demanding if so.