Moonshot has launched Kimi K3.1 Pro on the Kimi API, offering developers a new model option priced at $2.00 per million input tokens and $10.00 per million output tokens. Simon Willison reported on the release, noting that the model is currently closed weights, though Moonshot has committed to releasing full model weights by November 4, 2026. In addition to the standard release, a Highspeed version is available that runs two to three times faster at double the price, though input token cache hits retain standard pricing. According to Moonshot’s internal benchmarks, Kimi K3.1 Pro performs roughly on par with Claude Opus 5.5 and GPT-6 Astra, occasionally outperforming them on specific benchmarks. US dollar pricing for the model matches GPT-6.1 Sol while undercutting Claude Opus 5.5 by half.
Competitive Pricing and Benchmark Claims
The pricing strategy places Moonshot directly alongside established market competitors. Matching the input and output rates of GPT-6.1 Sol while offering a significant discount compared to Claude Opus 5.5 indicates an aggressive push for developer adoption. Moonshot’s benchmark claims suggest that the model can hold its own against top-tier alternatives from other major labs, giving developers a potentially cost-effective option for high-volume API workloads, particularly with the inclusion of the faster tier.
Unknowns and Future Open Weights Timeline
Despite the concrete API launch and pricing details, several important questions remain unanswered. The announcement did not specify the exact parameter count, architecture details, or infrastructure requirements for Kimi K3.1 Pro. Furthermore, the wait for open weights is substantial, with the promised release date set more than a year out in late 2026. Observers will need to monitor how adoption fares while the model remains closed and whether the Highspeed variant’s pricing model impacts overall cost efficiency in production environments.
Developers can access Kimi K3.1 Pro via the Kimi API immediately, while the broader community awaits further technical documentation and the eventual open weight release.
Lead image: panumas nikhomkhai / Pexels
