Moonshot Releases Kimi K3.1 Pro, With Open Weights Due November 4

Moonshot AI released Kimi K3.1 Pro on Friday, and it’s already in the chat apps, the coding tool and the API, with a context window of 1 million tokens.

If you build with AI, the bigger news is the promise attached: the full model weights are due by November 4.

Where you can use it now

Moonshot’s launch post on the Kimi blog lists where to try it today, so you can test it yourself:

• Kimi.ai
• Kimi Work
• Kimi Code
• the Kimi API, with K3.1 Pro and K3.1 Pro Highspeed

Moonshot says the Highspeed version runs two to three times faster.

K3.1 Pro itself is a post-trained upgrade of Kimi K3, and Moonshot calls it a 3T-class model.

Moonshot’s launch post also says it’s holding back cyber capability on purpose.

For the open-weight release, it found and removed cyber-related data from both training stages to reduce misuse risk. It kept the general coding tasks the model needs.

How it compares

CGTN’s report on the launch sets the price at $2 per million input tokens and $10 per million output tokens. That’s half of what Claude Opus 5.5 charges you.

So how close is it to the top? Here are the Terminal-Bench 4.0 coding agent scores: Claude Opus 5.5 at 66.4%, Kimi K3.1 Pro at 65.7% and GPT-6 Astra at 58.2%.

That’s less than a point behind Claude Opus 5.5 on the scores you just saw.

Meanwhile, Artificial Analysis ranked K3.1 Pro fourth on its Intelligence Index with a score of 55. Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full sit above it.

But you shouldn’t take those numbers at face value.

CGTN, citing OrcaRouter, notes that the Terminal-Bench scores were produced at different reasoning settings.

Your price edge is per token, so your task costs depend on usage. Artificial Analysis found K3.1 Pro uses about 55 thousand output tokens per task, 14% more than K3.

Why it matters

My read: open weights this close to the top of the leaderboard put pressure on every closed model’s price, and you’ll benefit even if you never touch Kimi.

Still, be realistic about self-hosting. For most of you, the API is the practical way in.

So where does it fit in your work? I’d try it on long jobs where a context window of 1 million tokens earns its keep:

• auditing your content archive in one pass
• rewriting your product pages
• reading a whole codebase before you change a plugin

That said, check your data rules first.

Do your clients care where their content gets processed? If they do, don’t switch provider before you ask them.

Then there’s the wait. If the full model weights land by November 4 with a usable license, your hosting company may offer it too, and you’ll have options.

Common Questions

When will Kimi K3.1 Pro’s open weights be released?

The full model weights are due by November 4, Moonshot says. The license hadn’t been announced at launch, so check the terms before you build on it.

How much does the Kimi K3.1 Pro API cost?

CGTN reports $2 per million input tokens and $10 per million output tokens. That’s roughly half what Claude Opus 5.5 costs, though your bill for a finished task depends on how many tokens the model uses along the way.

Can I use Kimi K3.1 Pro for coding?

Yes. Kimi Code and the Kimi API both give you access, and Moonshot’s own Terminal-Bench 4.0 score puts it close to Claude Opus 5.5.