Moonshot AI's Kimi K3 Arrives as the World's Most Advanced Open Model, at a Fraction of the Price
Moonshot AI launched Kimi K3 on July 16, 2026, a 2.8 trillion parameter model that ranked fourth on the Artificial Analysis Intelligence Index, priced well below Western competitors, with open weights expected by July 27.
The launch and the specs
Chinese AI company Moonshot AI launched Kimi K3 on July 16, 2026, a multimodal reasoning model with 2.8 trillion total parameters, a 1 million token context window and an always on thinking mode, according to technical breakdowns published by Graphify Guides and CodersEra. The model's API is available from launch, priced at $3 per million input tokens and $15 per million output tokens, with a reduced $0.30 per million tokens rate for cache hit input.
How a 2.8 trillion parameter model gets cheap to run
Kimi K3 uses a mixture of experts (MoE) architecture, with 896 experts in total, but activates only 16 of them per processed token, roughly 1.8% of the pool, according to the same technical documentation. That design is what lets a model this size cost a fraction of what it would if every parameter fired on every calculation, and explains why Kimi K3's per token price sits so far below comparably sized competitors.
Where the model lands on the intelligence ranking
On the Artificial Analysis Intelligence Index v4.1, an independent benchmark used to compare AI models, Kimi K3 scored 57.1, placing fourth overall, behind only Claude Fable 5 (59.9) and GPT-5.6 Sol (58.9), and ahead of Claude Opus 4.8 (55.7), according to data aggregated by LM Market Cap and the specialized site The AI Rankings. TrendingTopics went as far as calling the launch a China shocker that lifted open weight models to the industry's frontier level.
Open weights coming by July 27
Moonshot AI has promised to release the model's full weights by July 27, 2026, under a modified MIT license, which would make Kimi K3 the first open weight model in the 3 trillion parameter class available for download and fine tuning by any developer, according to CodersEra.
What it means for buyers of tokens
For SMBs and agencies running AI agents in production, Kimi K3 reinforces a trend already set by earlier Chinese releases: frontier, or near frontier, models becoming available at prices well below what Western labs charge, and soon with open weights for anyone who wants to run them on their own infrastructure. That widens the options for anyone currently picking a model on price per token alone, but it also raises the usual question: it is worth validating answer quality and data policy before switching providers just because of price.