Kimi unveils K3, an open-weight multimodal model competing with GPT-5.6 and Claude 3.5
Kimi introduces K3, a 2.8-trillion-parameter open-weight multimodal model with 1 million tokens of context, claiming performance close to GPT-5.6 Sol and Claude 3.5 Fable in internal benchmarks.
Published 17h1 sourceNotable
Lire en français
≈ 34s
The fact
K3 outperforms Opus 4.8 and GLM 5.2 in some evaluations, highlighting China's progress in closing the gap with Western AI leaders.
The model, priced higher than its predecessor, will have its full weights released by July 27, suggesting an end to ultra-low-cost Chinese AI offerings.
Click the link to read an article on the topic: