Kimi Releases K3.1 Pro With a 1 Million Token Context

Moonshot AI is a Chinese AI research group best known for its Kimi chatbot and coding tools. It has released a new model called Kimi K3.1 Pro. This model has a context window of 1 million tokens (the amount of text the model can process at once). It scored 65.7% on Terminal-Bench 4.0, a benchmark that measures how well models handle various terminal-related tasks. That is slightly below the 66.4% scored by Opus 5.5 (Claude). Still, it beats GPT-6 Astra, which scored 58.2%. Moonshot AI also ran its own internal test on a business-operations task. There, K3.1 Pro showed roughly a 4.2x improvement over the company's previous model, K3.

Based on these numbers, Kimi looks very close to matching Opus 5.5. We see this pattern often: Chinese models post strong benchmark scores, but fall short once people actually use them for real tasks. Most people will not get access to a model this capable anyway, even if Moonshot AI follows through on its plan to release open weights (model weights that anyone can download).