Chinese AI model Kimi K3.1 Pro now ranks fourth in the world

A Chinese AI model now ranks fourth in the world in one independent ranking. Kimi K3.1 Pro, from the Chinese lab Moonshot, trails only models from Anthropic and Meta. Free-to-download versions are due by Nov. 4, which would make it the strongest open model yet.

What changed

K3.1 Pro upgrades Moonshot's July K3 model. Artificial Analysis, an outside group that tests AI, placed it No. 4 on its Intelligence Index. It beat OpenAI's GPT-6 Astra. Only Anthropic's two 5.5 models and Meta's Muse Wonder ranked higher.

Moonshot also ran its own tests. It says K3.1 Pro beats Anthropic's Opus 5.5 on long coding projects, expert-level academic questions, research across many documents and real-world job tasks. Those results come from the company, not an outside tester.

What it costs

Pricing is $2 per million tokens in and $10 per million out. Tokens are the small chunks of text an AI reads and writes. That matches the price of Sonnet 5.5 and is about a fifth of Astra's rate.

There is a catch. K3.1 Pro uses more tokens than rival top models to do its work. The real cost of a task may be higher than the price tag suggests.

The security side

Artificial Analysis also calls K3.1 Pro the best openly available model for cyber defense. It scored 64% on the test and never refused a task for safety reasons. Only a GPT-6 variant, limited to vetted defenders, scored higher.

That matters because U.S. labs have kept their best hacking-related models locked down. The Rundown AI notes that Anthropic's Dario Amodei said in May that defenders had 6 to 12 months before Chinese models caught up. It adds that attack skills in open models are likely not far behind.

What to do

• If your business pays for AI, expect prices to keep falling as strong models get easier to get.
• Compare total cost per task, not just the price per million tokens.
• Keep software, passwords and security tools up to date, since capable AI is now available to more people.