Moonshot AI might’ve set the cat among the pigeons in the AI space.
The company has said that its latest model, Kimi K3, is behind only Fable and GPT 5.6 Sol in performance. “In our evaluations, Kimi K3 delivers frontier-level performance. Among the models tested, its overall intelligence ranks second only to Claude Fable 5 and GPT-5.6 Sol,” the company says in the Kimi K3 quickstart guide.
“Kimi K3 is Kimi’s most capable model to date, with 2.8 trillion parameters. Built on Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, it offers native visual understanding and a 1M-token context window for frontier intelligence scenarios such as software engineering, knowledge work, and deep reasoning,” the guide said. It added that Kimi K3 was the first open-source model to reach the 2.8-trillion-parameter scale.
Kimi K3 Architecture
“Kimi K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes),” the blog said. “Both architectural updates are designed to help information flow more smoothly through longer sequences and deeper models. We also further increased the sparsity of the Mixture of Experts (MoE): with the Stable LatentMoE framework, the model efficiently activates 16 out of 896 experts. Together with improvements in training methodology and data recipes, these structural advances give K3 roughly 2.5x the overall scaling efficiency of K2, converting compute into capability more effectively,” it added.
What will be interesting if Kimi K3 does really land just behind Fable 5 and GPT 5.6 Sol in benchmarks. It will put renewed pressure on OpenAI and Anthropic, both of which are looking to go public — it would mean that a very capable model is almost at the frontier, and is open-source to boot. Kimi K3 could significantly erode their pricing power, and customers — many of whom are now complaining about their AI costs — could be incentivized to switch. Also, it would put even further pressure on Google. Google had dropped out of the top 5 AI providers on the Artificial Analysis Intelligence Index, and Kimi K3 would be the second Chinese model after Z.AI’s GLM 5.2 to eclipse it. It remains to be seen what Kimi K3’s official benchmarks are, but if Moonshot AI’s internal evaluations are correct, the release might end up shaking up the AI space in a big way.