Kimi K3.1 Pro costs half as much as Claude Opus 5.5, with open weights promised by November 4

Moonshot AI released Kimi K3.1 Pro on its API on October 23. Developer Simon Willison reports that it costs $2.00 per million input tokens and $10 per million output tokens. That matches GPT-6.1 Sol and is half the price of Claude Opus 5.5.

A token is a small chunk of text, roughly a short word, and API providers bill by how many go in and come out. Moonshot also sells a “Highspeed” version that is 2 to 3 times faster at twice the price. Cached input, meaning text the model has already seen in a recent request, costs the same on both versions.

Open weights are promised, not released

Moonshot says “Full model weights will be released by Nov 4, 2026.” Until then, Moonshot offers hosted access through Kimi, Kimi Work, Kimi Code and its API. Open weights are the model files themselves, which let you or another host run the model on your own servers. Check the license when the files appear, since its terms decide whether you can use the model commercially or self-host it.

The new model follows Kimi K3, which launched in July at $3 per million input tokens and $15 for output.

Benchmarks put it near the top, with a gap

Moonshot’s own benchmarks show K3.1 Pro roughly level with Claude Opus 5.5 and GPT-6 Astra. It beats them on a few tests and comes close on others.

The independent score is more useful. In his October 23 report, Willison cites a score of 55 from Artificial Analysis, an independent benchmarking service. That is ahead of Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon at 53 each, and behind only Claude Opus 5.5 at 58. Willison calls it “the first time a (soon-to-be) open weight model has scored this highly against the frontier.”

A single index score summarizes many kinds of tasks. It does not show how the model handles your codebase, your tools or your tolerance for slow responses.

Compare cost per finished task, not per token

K3.1 Pro scores within three points of Opus 5.5 on this index at half the per-token price. If you run long coding-agent sessions on Opus 5.5, that gap could show up on your bill.

Per-token price is only part of the cost, because a model that thinks longer or needs more retries uses more tokens. Compare the cost of a completed task, including corrections, before expecting the lower rate to reduce your bill.

If Moonshot releases the weights on November 4 as promised, builders will gain the option of running the model independently. The license and the hardware needed to serve it will determine whether that option is practical for your project.

End of article