Introducing Kimi K3.1 Pro

Kimi K3.1 Pro has made its debut on the Kimi API, signaling a significant step forward in open weight model capabilities. Despite not being open weights at launch, the model carries the promise of Moonshot capabilities. The developers have announced that full model weights will be released by November 4, 2026.

In addition to the standard version, Kimi K3.1 Pro offers a Highspeed variant, which is notably 2-3 times faster at twice the price. The pricing remains comparable to that of GPT-6.1 Sol, with input priced at $2.00 per million tokens and output at $10 per million tokens. This pricing is half the cost of Claude Opus 5.5, making it an attractive option for those seeking high performance without breaking the bank.

According to benchmark results, Kimi K3.1 Pro matches the performance of Opus 5.5 and GPT-6 Astra in certain tests, and performs close to these models on others. The Artificial Analysis gave Kimi a score of 55, which places it ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon, all of which scored 53. Kimi K3.1 Pro falls just behind Claude Opus 5.5, which scored 58, but this is a first for an upcoming open weight model to achieve such a high ranking.

The author tested Kimi K3.1 Pro using the Moonshot API via the llm openai endpoint command, which was added to the LLM 0.32 in August. This command allows prompts to be run against any OpenAI-compatible endpoint, without logging the results. The author provided examples of the model's output at three reasoning levels - low, medium, and max. The max reasoning level output was particularly impressive, serving as a point of comparison to Kimi K3's performance at the same levels.