Kimi K3.1 Pro: a Chinese model almost at the top, and soon open to everyone

China’s Moonshot AI has released Kimi K3.1 Pro. According to the testing firm Artificial Analysis it is the fourth best AI model right now, only a few points behind Claude Opus 5.5, and its weights are due to be free to download by 4 November. Here is what it means for AI prices, and what to think about before you try it.

On 23 October 2026 Moonshot AI released Kimi K3.1 Pro and a faster variant, K3.1 Pro Highspeed, which the company says is two to three times faster. It is already available on kimi.ai, in the Kimi mobile apps, in the desktop app Kimi Work, in the coding tool Kimi Code and through the Kimi API. Moonshot promises to release the model weights, that is the model itself, by 4 November 2026. Then anyone can download it and run it on their own servers.

How good is it? The independent testing firm Artificial Analysis gives it 55 points on its Intelligence Index. That puts it in fourth place, behind Claude Opus 5.5 with 58 and Claude Sonnet 5.5 and Muse Wonder Full with 56 each. Once the weights are released it will, according to Artificial Analysis, be the best open model, and the gap between the best open and closed models shrinks from 12 to 3 points. The model is large: 2.8 trillion parameters in total, of which 104 billion are used for each word, and it can read up to one million tokens, roughly several thick books, at once.

The price is what stands out. Through the API, K3.1 Pro costs 2 dollars per million input tokens and 10 dollars per million output tokens, a third cheaper than its predecessor Kimi K3. Highspeed costs twice as much. In Kimi’s own apps the model is available with limited usage on the free plan, and paid plans start at 15 dollars a month billed annually. A nearly equally good model becoming open puts pressure on prices even for the big American services.

Moonshot writes that it deliberately removed training data about cyberattacks for the open version, so that it does not become too good at hacking. In the company’s own test, the final model solved 81.0 percent of the tasks, against 90.5 percent for a version without that restriction. These are Moonshot’s own figures, as is the result on the coding test Terminal-Bench 4.0, where Moonshot reports 65.7 percent against 66.4 for Claude Opus 5.5.

Good to know: Moonshot AI is a Chinese company. Read the privacy terms before you enter anything personal or sensitive in Kimi’s apps, and do not use it for customer data at work unless it is approved. The open weights are a different matter: then a company can run the model with a provider it chooses itself. But the model is too large for an ordinary computer and needs powerful servers.

Sources

• Moonshot AI: Introducing Kimi K3.1 Pro (teknisk blogg, tillgänglighet och vikter senast 4 nov 2026) (retrieved 2026-10-25)
• Kimi: Kimi K3.1 Pro Pricing – Membership Plans and API Costs (uppdaterad 24 okt 2026) (retrieved 2026-10-25)
• Artificial Analysis: Kimi K3.1 Pro launches at 4th on the Intelligence Index (23 okt 2026) (retrieved 2026-10-25)

TipAI explains AI in plain language for beginners, families and small businesses. Every article has sources and a "How to get started" section.

← All news · Try TipAI