Moonshot AI

Artificial Intelligence · Open-Weight Models · Foundation Models

Launch

Moonshot launches Kimi K3.1 Pro, first big upgrade since DeepSeek moment

The open model now trails Anthropic's best by under a point on coding tests, narrowing the gap with closed-source leaders.

• Moonshot AI released Kimi K3.1 Pro on October 23, 2026, its first major upgrade to Kimi K3 since that model's July debut, with full open weights due by November 4.
• Beijing-based Moonshot AI is the $50 billion startup whose July release of Kimi K3 drew comparisons to DeepSeek's early-2025 breakthrough in the West.
• K3.1 Pro is a mixture-of-experts model with 2.8 trillion total parameters, 104 billion active at once, a 1-million-token context window, and pricing cut 33% versus K3.
• On the Terminal-Bench 4.0 coding test, Moonshot says K3.1 Pro scored 65.7%, beating GPT-6 Astra's 58.2% and landing within one point of Claude Opus 5.5's 66.4%.
• Moonshot says it stripped cyberattack-related data from training before the open release, which cut its internal exploit-benchmark score to 81% from 90.5% without those safety curbs.
• The launch lands as Chinese models top global usage charts, with China's weekly AI model calls surpassing the US for 25 straight weeks and claiming the top five spots on OpenRouter.
• Open-weight Chinese labs are now closing in on closed-source frontier models within single-digit benchmark points, pressuring US labs to justify premium pricing for proprietary access.