Kimi K3.1 Pro: Open Weights, Almost Opus

Moonshot just released Kimi K3.1 Pro, an open-weight mixture of experts model with 2.8 trillion total parameters, about 104 billion active, and a 1 million token context window. It’s the same size as Kimi K3, post-trained with a lot more reinforcement learning. On Moonshot’s own Terminal-Bench 4.0 numbers it scores 65.7%, less than a point behind Claude Opus 5.5 (66.4%) and ahead of GPT-6 Astra (58.2%), at about a third of Opus’s cost per task on their chart. Moonshot also removed cyber-related data from training before the open release. Even so, it solves 81% of their internal exploit tasks against 33% for Kimi K3. It’s live today on the Kimi API at $2 input / $10 output per million tokens, and the open weights are due by November 4, 2026. All benchmark numbers are Moonshot’s own; independent results aren’t in yet. Moonshot tech blog: https://www.kimi.ai/blog/kimi-k3-1-pro Launch thread: https://x.com/Kimi_Moonshot/status/2113680701770068067 Kimi API: https://platform.kimi.ai human-atlas (the walking-sim demo’s base): https://github.com/ashemag/human-atlas #KimiK3 #MoonshotAI #OpenWeights #AI #Shorts