Kimi K3.1 Pro: pricing, benchmarks and weights release
Moonshot’s Kimi K3.1 Pro has a one-million-token context. API pricing, vendor-reported benchmarks and full weights promised by 4 November 2026.
Checked on 26 October 2026 against the linked official sources. Prices and access can change; evaluation results are attributed to the vendor. The full weights remain a future commitment, announced by 4 November.
Direct answer: Kimi K3.1 Pro is Moonshot AI’s new K3.1 model with a one-million-token context window. Released on 23 October 2026, it is available through Kimi products and the API. Moonshot promises the full weights by 4 November 2026; they are not yet available for local download at this article’s cutoff.
It is a post-trained version of Kimi K3. Moonshot describes it as a 3T-class model and introduces a Highspeed variant claimed to run two to three times faster.
Availability and API prices
Access is available in Kimi.ai and mobile apps, Kimi Work, Kimi Code and the Kimi API. The API identifiers are kimi-k3.1-pro and kimi-k3.1-pro-highspeed.
- K3.1 Pro: USD 2 per million input tokens and USD 10 per million output tokens.
- Highspeed: USD 4 input and USD 20 output per million tokens.
These are the standard input/output prices on the API platform checked on 26 October. Cache pricing is listed separately. The announced weights date is a future commitment, rather than a completed release.
Benchmarks reported by Moonshot
On Terminal-Bench 4.0, Moonshot reports 65.7% for K3.1 Pro, compared with 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5 in its comparison. On its internal Kimi Code Bench 3.0 Long-horizon, it reports 61.2%, versus 55.95% for K3.
The company also reports a drop in claim-level hallucinations from 23.4% to 8.3% on Kimi Hallu Bench Hard. That deliberately difficult internal test is not an estimate of everyday error rates.
These figures are vendor-reported. They are not a controlled comparison carried out for this article or proof that the model will outperform alternatives in a particular project.
Coding, 3D work and financial workflows
Moonshot highlights full-project coding in game engines, parametric 3D work using tools such as Blender and CAD, and financial research and modeling. Treat launch examples as candidates for your own trial: provide the same brief, tool access and acceptance criteria to each model.
Cyber capability controls and open weights
Moonshot says it removed cyber-related data during mid-training and reinforcement learning to reduce misuse risk. Its internal xdev-bench-hard comparison reports 81.0% pass@3 for the released model and 90.5% without those measures.
The practical questions still open are the actual weights release, license and deployment requirements, plus the hardware needed to run this large model. Do not assume a consumer computer can run it simply because weights are promised.
Frequently asked questions
When was Kimi K3.1 Pro released? On 23 October 2026, through Kimi products and the API.
When will the weights be available? Moonshot promises the full weights by 4 November 2026. This is still a future release at the article’s cutoff.
How much does the API cost? Standard K3.1 Pro input/output costs are USD 2/10 per million tokens; Highspeed costs USD 4/20. Cache pricing is separate.
Is it better than Opus 5.5 or GPT-6 Astra? Moonshot reports 65.7% on Terminal-Bench 4.0, against 66.4% for Opus 5.5 and 58.2% for Astra in its comparison. Your own task results may differ.
Can I run it locally now? The full weights are not yet available at this article’s cutoff. Access is through Kimi products and its API.