Kimi K3: A New Frontier for Open Source AI

Chinese AI company Moonshot AI has released Kimi K3, a 2.8 trillion parameter model that its benchmark scores suggest rivals—and in some areas exceeds—the best proprietary models from OpenAI and Anthropic. The model is open source, with full weights scheduled for release by July 27th, and available now for free at kimi.com with no credit card required. The release lands alongside news that Google has fallen months behind schedule on its Gemini 3.5 Pro flagship, and that Major League Baseball has banned dugout iPads from accessing generative AI for in-game decisions after discovering roughly a third of teams were already using it that way.

Kimi K3 Closes the Open Source Gap

The headline numbers are striking. Kimi K3 scored 88.3 on Terminal Bench 2.1, trailing only GPT 5.6 Sol's 88.8. On GDP Val Double A V2, a benchmark measuring real-world tasks across 44 occupations and nine industries, K3 scored 1687—third behind Claude Fable 5 Max at 1815 and GPT 5.6 Sol Max at 1747.8, but ahead of Claude Opus 4.8 at 1600. On BrowseComp, a benchmark for long-horizon information seeking, K3 achieved a state-of-the-art score of 91.2 out of 100, accomplished in a single-agent setup using its 1 million token context window without any context compression or additional management techniques.