Published: July 16, 2026
Moonshot AI has released Kimi K3, a 2.8-trillion-parameter open-weight multimodal reasoning model with a context window of roughly 1 million tokens. Kimi K3 ranks fourth of 189 models on the Artificial Analysis Intelligence Index, on par with Claude Opus 4.8 and GPT-5.5. The model went live through Kimi’s apps and API on July 16, 2026, with full open weights scheduled to follow by July 27.
What is Kimi K3?
Kimi K3 is an open-weight multimodal reasoning model from Moonshot AI, the Chinese lab behind the Kimi assistant. It uses a mixture-of-experts (MoE) design with 896 experts, of which 16 are active per token, and a total of 2.8 trillion parameters. Tom’s Hardware describes it as the largest open-weight AI model released to date.
The model handles a context window of 1,048,576 tokens, roughly one million, which allows it to process very long documents, codebases, or multi-step agent sessions in a single pass. It became available through Kimi’s apps and API on July 16, 2026, ahead of the full open-weights release scheduled for July 27.
Kimi K3 benchmarks and technical specs
On the Artificial Analysis Intelligence Index, Kimi K3 scores 57 and ranks fourth out of 189 models, level with Claude Opus 4.8 and GPT-5.5 and behind only Claude Fable 5 and GPT-5.6 Sol. In the Frontend Code evaluation it ranks first with 1,679 points, ahead of Claude Fable 5.
The 2.8-trillion-parameter mixture-of-experts design keeps only 16 of 896 experts active per token, which limits the compute cost of each query despite the very large parameter count. Combined with the roughly 1 million-token context window, this positions Kimi K3 for long-context reasoning and coding tasks.
How much does Kimi K3 cost?
Through the Kimi API, Kimi K3 is priced at $3 per million input tokens and $15 per million output tokens. Cached input is billed at $0.30 per million tokens, flat across the full context window.
Because the weights are scheduled to become openly available by July 27, 2026, teams will also be able to run Kimi K3 on their own infrastructure rather than only through the hosted API. That combination of frontier-level ranking and open weights is the main reason the release drew wide attention.
How does Kimi K3 compare to Claude Opus 4.8 and GPT-5.5?
On the Artificial Analysis Intelligence Index, Kimi K3’s score of 57 places it on par with Claude Opus 4.8 and GPT-5.5, both proprietary models, while Kimi K3 ships as open weights. It leads the Frontend Code evaluation at 1,679 points, ahead of Claude Fable 5.
The significance is less about topping every benchmark and more about an open-weight model reaching the same tier as leading closed models. For organisations that need to self-host, Kimi K3 narrows the gap between open and proprietary systems.
Kimi K3 is Moonshot AI’s 2.8-trillion-parameter open-weight model with a roughly 1 million-token context window and a fourth-place ranking on the Artificial Analysis Intelligence Index.
Full specifications are available through Moonshot AI’s Kimi K3 documentation, with open weights due by July 27, 2026.
