Moonshot's Kimi K3.1 Pro: half Opus's price

Moonshot AI's new Kimi K3.1 Pro costs half as much as Anthropic's Claude Opus 5.5, the two price pages show. On Moonshot's own coding test it scored 0.7 points lower.

Moonshot released it this week on Kimi.com, its phone apps, Kimi Work, Kimi Code and its API, the company's blog says. MarkTechPost reported the launch on the 23rd of October 2026.

Open weights by the 4th of November, with 1 million tokens of memory

Moonshot calls it its most capable open-weight model, one anyone can download and run. The full weights come by the 4th of November 2026, the blog says.

It reads up to 1 million tokens at once, the blog says. Tokens are small pieces of words.

A faster Highspeed version runs 2 to 3 times quicker, Moonshot wrote.

API prices: 2 dollars in and 10 out, against Opus 5.5's 4 and 20

Kimi K3.1 Pro costs 2 dollars per million tokens read and 10 dollars per million written, Moonshot's price page shows. Claude Opus 5.5 costs 4 and 20 dollars, Anthropic's price page shows.

Below we break down what that means for one big job.

Reading 1 million tokens and writing 1 million costs 12 dollars on Kimi K3.1 Pro and 24 on Opus 5.5, by our arithmetic. The older Kimi K3 costs 18 dollars for the same job.

What is confirmed and what is not

On Terminal-Bench 4.0, a test of coding work in a terminal, Kimi K3.1 Pro scored 65.7 percent, Moonshot wrote. Opus 5.5 scored 66.4 and OpenAI's GPT-6 Astra 58.2.

These are Moonshot's own runs.

Moonshot also wrote that it removed cyber attack data from training before the open release. Its stated aim is to keep open weights behind the strongest closed models on hacking skills.

Key facts

Released by Moonshot AI this week, reported on the 23rd of October 2026.

Open weights due by the 4th of November 2026.

Price: 2 dollars in, 10 dollars out per million tokens.

Half the price of Claude Opus 5.5.

Terminal-Bench 4.0: 65.7 percent, Moonshot's own test.