Moonshot AI releases Kimi K3.1 Pro with Highspeed variant

Moonshot AI on Friday, October 23, released Kimi K3.1 Pro, the first model in its new K3.1 series. It also released Kimi K3.1 Pro Highspeed, which the company says runs two to three times faster. In its technical blog, Moonshot calls K3.1 Pro its "most capable 3T-class open-weight model to date," with a 1-million-token context window. The company says the full weights will be released by November 4. Bloomberg called the launch the first major upgrade to the model that pushed the Beijing-based startup, now valued at $50 billion, into the frontier of AI development.bloomberg.com+1

Benchmarks and focus areas

Moonshot says the new model is much better at financial research and coding, two areas US rivals are also targeting, and more capable at 3D modeling for tasks such as game design, Bloomberg reported. On Terminal-Bench 4.0, a test of coding agents on terminal tasks, the company reports a score of 65.7%. That puts it ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%, according to Moonshot.kimi.ai+1

The company also reports a score of 92.7% with tools on the BenchCAD 3D reconstruction test and 64.4 on the OfficeQA Pro v2 financial-document test, Unite.AI reported. On an internal stress test for factual errors, the rate of hallucinated claims fell from 23.4% for K3 to 8.3%. Moonshot cautions that this test was designed to probe error-prone situations and does not reflect error rates in everyday use. K3.1 Pro is a post-trained version of Kimi K3 built with much more post-training compute. For the open-weight release, Moonshot says it removed cyber-related training data so the model would not push the frontier of offensive cyber capability.unite.ai

AlphaSignal describes the model as a mixture-of-experts system with 2.8 trillion total and 104 billion active parameters. It notes that the license for the downloadable weights has not been confirmed.alphasignal.ai

Pricing tiers in Kimi Code

Both models are available now on Kimi.ai, Kimi Work, Kimi Code and the Kimi API, Moonshot said. The Kimi Code documentation says the Highspeed version has the same capabilities as K3.1 Pro but uses about twice the quota, RuntimeWire reported. Both models run only in thinking mode. Under the new membership plans, K3.1 Pro requires the Plus tier or higher, while the Highspeed version requires Pro or higher. The 1-million-token context window is limited to Pro members and above, according to Unite.AI.runtimewire.com+2

A teaser-heavy run-up

The launch came after weeks of leaks and an anonymous model called "Kibitz" on the Arena testing site, which testers believed was K3.1, TestingCatalog reported. Early on October 23, Moonshot's official account posted a 24-second video with no caption showing "Kimi K3.1" running in Kimi Code, PC Watch reported. Kimi K3 came out in July; its API launched on July 17 and its weights followed about 10 days later, according to BlockTempo. Moonshot's new benchmark claims have yet to be tested independently. As AlphaSignal noted, the model's value will depend on controlled evaluations and the final license.pc.watch.impress.co.jp+3