Kimi K3.1 Pro launched on the Moonshot API today, with full model weights promised by 4 November 2026. The service offers a Highspeed version running at two to three times the normal rate for double the cost, excluding input token cache hits. Pricing stands at two dollars per million input tokens and ten dollars per million output tokens, matching GPT-6.1 Sol and costing half of Claude Opus 5.5. Benchmarks place the model level with Opus 5.5 and GPT-6 Astra, while Artificial Analysis scored it at 55, ahead of Gemini 4 Argon and Claude Fable 5.1. This marks the first instance where a soon-to-be open weight model scores this highly against the frontier.
The practical implication is that Chinese models now compete directly with western leaders on standardised tests. Moonshot provides a highspeed option for latency-sensitive tasks, though the price increase limits its appeal for heavy inference workloads. The release also signals a shift towards open weights as a standard for top-tier models.
- Full model weights release date: 4 November 2026
- Highspeed tier is 2-3x faster at double the price
- Artificial Analysis score: 55
