Introducing Kimi K3.1 Pro
Simon Willison’s Weblog Simon Willison
Kimi K3.1 Pro is on Moonshot’s API now, with open weights promised later. It’s priced like a serious frontier model and is already posting benchmark numbers right up there with the best.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Moonshot has put Kimi K3.1 Pro on the Kimi API, but not as open weights yet. The company says the full weights will arrive by Nov. 4, 2026. For now, there’s also a “Highspeed” version that runs 2-3x faster and costs twice as much, with one catch: input token cache hits stay at the same price.
The standard price in US dollars is $2.00 per million input tokens and $10 per million output tokens. That matches GPT-6.1 Sol and comes in at half the price of Claude Opus 5.5. So this is not being sold as a bargain-bin model. It’s being positioned as something much closer to the top tier.
Moonshot’s own numbers put Kimi roughly alongside Opus 5.5 and GPT-6 Astra. On some tests it edges them out, on others it lands near them. Artificial Analysis gave it 55, ahead of Claude Fable 5.1, GPT-6 Astra, and Gemini 4 Argon, all on 53, and just behind Claude Opus 5.5 on 58.
That last part is the eyebrow-raiser. This is the first time a model that’s expected to be open weight has scored this high against the frontier, at least in this set of rankings. The gap between “open” and “top-tier” keeps getting smaller, and Moonshot has just made that much harder to ignore.
Simon Willison also tried it through the Moonshot API using LLM’s OpenAI-compatible endpoint, which he’d added to LLM 0.32 in August. He tested Kimi K3.1 Pro with prompts like “Generate an SVG of a pelican riding a bicycle,” and compared the model across the low, medium, and max reasoning levels it supports. He says the results are worth looking at, and the comparison with Kimi K3 makes the point even more clearly.
My take — AI-written commentary, not fact-checked reporting
The real story here isn’t “open weights someday”; it’s that frontier-grade models are now being marketed with a delayed freedom date like software licenses on layaway. That’s a very modern kind of tease. Open-weight believers should enjoy the benchmark win, but the industry still knows how to keep a tight grip while sounding generous.
Read more about this at: Simon Willison’s Weblog
