Moonshot released Kimi K3.1 Pro, a 2.8 trillion parameter model with a 1M token context

Moonshot AI released Kimi K3.1 Pro on October 24, the first model in its new K3.1 series, along with a faster variant called Kimi K3.1 Pro Highspeed. The company describes K3.1 Pro as its most capable 3T-class open-weight model so far, with 2.8 trillion parameters and a context window of one million tokens.

K3.1 Pro is a post-trained version of Kimi K3, built with substantially more post-training compute and large scale reinforcement learning across hundreds of domains. Moonshot says the Highspeed variant runs two to three times faster with the same capabilities. Both are available on Kimi.ai, the Kimi mobile and desktop apps, Kimi Code and the Kimi API, where they are listed as kimi-k3.1-pro and kimi-k3.1-pro-highspeed, with zero data retention available for API customers. The Decoder reports that API prices drop by 33 percent compared with K3.

On performance, independent tracker Artificial Analysis puts K3.1 Pro at 55 points on its Intelligence Index, an 11 point jump over K3 and fourth place overall, behind only Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full. Moonshot's own charts show its strongest result on SWE-Marathon at 51.7 and a weaker 38.1 on Terminal-Bench Science. Those per-benchmark figures are self-reported.

The weights are not out yet. Moonshot plans to release them by November 4. If that happens, The Decoder notes, K3.1 Pro would become the highest scoring open weights model available and shrink the gap between the best open and proprietary models to about three points.

Why it matters for developers: a near frontier model with open weights changes the build versus buy math for teams that need to self-host for privacy or cost. At 2.8 trillion parameters, though, this is not a single GPU model. Running it yourself means multi-node serving, so most teams will meet it through the API or third party hosts first.

What's next: watch for the November 4 weights drop, the license terms that come with it, and independent agentic coding tests to see whether the self-reported numbers hold up.

Building with AI? See this week's hot AI repos and how to use each one.