Moonshot AI’s Kimi K3 narrows the frontier gap

7/19/2026

Kimi K3 open-weights model rivals frontier AI with 1M context and strong benchmarks. See how Moonshot prices it near Claude 5 Sonnet before weights land July 27.

Moonshot AI has released Kimi K3, an openweights model from Beijing that the company says performs near the level of leading frontier systems such as Anthropic’s Claude Fable 5 and OpenAI’s GPT5.6 Sol on several benchmarks. According to the report, Kimi K3 reaches a 1 million token context window and shows strong results in areas including web research, spreadsheets, frontend design, and long coding tasks. It also scored 57 on Artificial Analysis’ Intelligence Index, placing it just behind the top proprietary models highlighted in the article. Moonshot says the model will be available at pricing similar to Claude 5 Sonnet, and its weights are expected to be published by July 27. The release adds to the competitive pressure in the AI model market, especially as opensource and Chinese labs continue to close the gap with the most advanced closed models.