Moonshot's Kimi K3.1 Pro, its most capable model to date with 3 trillion parameters, officially launches
Moonshot has launched Kimi K3.1 Pro, the company's most capable model to date, with 3 trillion parameters and 1 million context length, alongside a faster Highspeed variant that runs 2–3x faster. On Terminal-Bench 4.0 it scored 65.7%, beating GPT-6 Astra's 58.2% and nearly matching Claude Opus 5.5's 66.4%. The model is already available via kimi.com, the app and API, with full weights promised before November 4. Why it matters: a Chinese flagship is now competitive with top closed models on agentic coding tasks, shifting the frontier for openly released models.
SOURCES
- IT之家2026-10-23 23:08