Chinese AI lab Moonshot AI shipped Kimi K3 this morning, describing it as their "most capable model to date" with 2.8 trillion parameters.

The model is available via the company's website and API, with open weights promised by July 27. Moonshot calls K3 the first "open 3T-class model", taking the crown from DeepSeek's 1.6T v4 Pro.

Self-reported benchmarks show K3 mostly beating Claude Opus 4.8 max and GPT-5.5 high, while losing to Claude Fable 5 and GPT-5.6 Sol. On Artificial Analysis's private long-horizon knowledge work evaluation, K3 reached an Elo of 1547, gaining 732 points from Kimi K2.6.

The model now leads Arena.ai's Frontend Code arena, surpassing even Claude Fable 5.

Premium pricing signals shift

K3 costs $3 per million input tokens and $15 per million output tokens. This matches Anthropic's Claude Sonnet series pricing and makes it the most expensive model from a Chinese AI lab.

The pricing represents a significant jump from earlier models like Kimi K2.6 at $0.95/$4. The 2.8 trillion parameter count is also more than double that 1T model's size.

Cost per task runs $0.94, similar to GPT-5.6 Sol at $1.04 but roughly half the price of Opus 4.8 at $1.80, according to Artificial Analysis. The model uses 21% fewer output tokens than K2.6.

K3 currently offers only one reasoning effort level, "max", which consumed 13,241 reasoning tokens to generate a simple SVG test. The model includes vision capabilities and accepts image input.

Simon Willison's pelican benchmark test revealed the model uses an estimated 85-token hidden system prompt, with a simple "hi" prompt counting 86 tokens total.

Moonshot plans to release open weights by July 27, which would make K3 the largest openly available model by parameter count.