Skip to content
What is Kimi K3.1 Pro? Moonshot AI's new flagship model
Back to Blog
#AI models#Kimi K3.1 Pro#kimi-k3.1-pro#Kimi K3.1 Pro pricing#Kimi K3.1 Pro API#Moonshot AI

What is Kimi K3.1 Pro? Moonshot AI's new flagship model

Amir Khan

Kimi K3.1 Pro is Moonshot AI's new flagship model, built for long coding jobs, agents and document-heavy work such as financial research. It went live on 23 October 2026 in the Kimi apps, Kimi Work, Kimi Code and the Kimi API, with a 1-million-token context window. Through the API it costs $2 per million input tokens and $10 per million output tokens, half the $4 and $20 Anthropic charges for Claude Opus 5.5. Moonshot says the full model weights will be released by 4 November 2026.

If you pay frontier prices for coding agents or long-document work, test it against the model you use now. Expect it to be slow and to spend a lot of tokens thinking. Budget on the tokens it actually uses, not on the price per token.

What changed from Kimi K3?

The price dropped and the training went further. Moonshot describes K3.1 Pro as a post-trained version of Kimi K3, trained with "substantially larger post-training compute". The dates and prices below come from Moonshot's announcement and API docs, read on 25 October 2026:

  • 23 October: kimi-k3.1-pro and a faster variant, kimi-k3.1-pro-highspeed, go live. Moonshot says Highspeed runs 2-3 times faster at twice the price.
  • Price: K3 costs $3 in and $15 out per million tokens. K3.1 Pro costs $2 and $10, and cached input drops from $0.30 to $0.20.
  • 31 October: new accounts can no longer use the K2 series. Existing accounts keep access.
  • 4 November: the deadline Moonshot gives for releasing the full weights.
  • End of December: Moonshot plans to retire all K2 models. If you still run K2, plan the move now. The docs say to remove the K2 thinking setting and use reasoning_effort instead.

How good is it?

On Moonshot's own figures it sits close to the best proprietary models. It scores 65.7% on Terminal-Bench 4.0, a test of coding agents working in a terminal. That puts it narrowly behind Claude Opus 5.5 at 66.4% and ahead of GPT-6 Astra at 58.2%.

Artificial Analysis, which runs its own tests, scores it 55 on its Intelligence Index. Of the models it compares there, only Claude Opus 5.5 ranks higher, at 58. The same tests show the costs. K3.1 Pro produced 73 output tokens a second, against an average of 82. Running the index took it 150 million tokens, against a median of 87 million.

Moonshot chart of Terminal-Bench 4.0 score against average cost per task for Kimi K3.1 Pro, Claude, GPT-6 Astra, GLM-5.3 and DeepSeek models
Terminal-Bench 4.0 score against cost per task, from Moonshot's K3.1 Pro announcement. Image: Moonshot AI.

Security teams should note one deliberate limit. Moonshot removed cyber-related training data from K3.1 Pro before releasing the weights. It says the model's offensive security skills remain behind those of the leading proprietary models.

What does a job cost?

Here is a code review that sends 40,000 tokens of code and gets back 8,000 tokens of reasoning and answer:

  • K3.1 Pro: 40,000 × $2 ÷ 1,000,000 = $0.08 input, plus 8,000 × $10 ÷ 1,000,000 = $0.08 output. Total: $0.16.
  • K3.1 Pro Highspeed, at $4 and $20: $0.32.
  • Claude Opus 5.5, at $4 and $20 on Anthropic's pricing page read on 25 October: $0.32.

The output figure is the one to watch. K3.1 Pro always reasons, its reasoning effort defaults to max, and Moonshot's docs confirm reasoning tokens are billed. Set reasoning_effort to low or high where the hardest setting is not needed, and measure real token use on your own prompts. Prices exclude tax.

How do you start?

Create a Kimi API Platform account, then top up at least $1, which unlocks the model. Your total top-ups set your rate limits. The API works with the OpenAI SDK. You point it at https://api.moonshot.ai/v1 and use the model id kimi-k3.1-pro. This is the basic call from Moonshot's quickstart, read on 25 October 2026:

curl https://api.moonshot.ai/v1/chat/completions \
  --header "Authorization: Bearer $MOONSHOT_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "kimi-k3.1-pro",
    "messages": [{"role": "user", "content": "Introduce Kimi K3.1 Pro in one sentence."}]
  }'

For multi-turn chats and tool calls, the docs say to send the complete assistant message back, including reasoning_content and tool_calls, not just the text. Zero Data Retention is available for API customers. On Kimi Enterprise, the model is off by default until an administrator turns it on.

Want Kimi K3.1 Pro built into your product or workflow, and tested against the model you use today? See what we build or tell us about the job.

Sources