Moonshot AI launched Kimi K3.1 Pro on October 23, making the model available across Kimi’s consumer apps, Kimi Work, Kimi Code and its API. The Beijing company calls it the first model in its K3.1 series and its most capable model in the three-trillion-parameter class.
A broad product launch before the weights arrive
The release includes a Kimi K3.1 Pro Highspeed option that Moonshot says delivers the same capabilities at two to three times the speed. Enterprise administrators can enable the model, while API customers can select standard and Highspeed variants. Moonshot says the model supports a one-million-token context window.
The company says full model weights will be released by November 4. That distinction matters: K3.1 Pro is available to use now, but the promised open-weight artifact was not yet available at announcement time.
Coding, 3D work and finance are the headline targets
Moonshot emphasizes coding, long-horizon agent work, three-dimensional creation and financial research. It reports gains over Kimi K3 across its internal tests and publishes comparisons with proprietary models, but many numbers use Moonshot-run harnesses or internal benchmarks and should be read as vendor evidence rather than neutral proof.
The launch is significant because it adds another Chinese model family to the top tier of globally distributed agentic systems while committing to a near-term weights release. Independent evaluator Artificial Analysis also ranked the model near the top of its current index, though rankings depend on each evaluator’s task mix and methodology.