Kimi K3.1 Pro
Moonshot's post-training upgrade to Kimi K3 focuses on sustained coding and agent work, with a million-token context window and stronger results on its long-horizon coding test.
Kimi K3.1 Pro is available on MiniRouter as . The model page has current pricing, provider routes and API setup.
More training for work that takes several steps
K3.1 Pro builds on Kimi K3 with a larger post-training run. Moonshot focuses the release on coding, 3D creation and document research: tasks where an agent has to act, inspect the result and keep going.
- Predecessor
- Kimi K3
- A post-training upgrade, and the first model in the K3.1 series.
- Context
- 1,048,576 tokens
- Input, reasoning and the answer share the context window.
- Reasoning
- Low · high · max
- The effort options currently listed in MiniRouter's catalog.
- Weights
- Planned by November 4
- Moonshot's announced release target; the weights are not yet available.
A stronger result on long coding tasks
Moonshot's internal Kimi Code Bench 3.0 Long-horizon tests multi-step execution, tool use and complete engineering tasks. K3.1 Pro improves on K3 by 5.25 percentage points.
Kimi Code Bench 3.0 Long-horizon
- K3.1 Pro61.2%
- K356.0%
0–100% · higher is better
View chart data
| Model | Effort | Score |
|---|---|---|
| Kimi K3.1 Pro | max | 61.2% |
| Kimi K3 | max | 55.95% |
For an existing agent, the useful comparison is a completed change: give both models the same repository task, tools and test suite, then compare correctness, cost and time.
From a prompt to something you can inspect
The release includes examples of game projects, Blender and CAD scripts, and financial research. These workflows depend on tools around the model. Your application provides the runtime, retrieves documents and executes the actions.
Describe the task
Provide source material, constraints and the tools your application exposes.
Execute tool calls
Your application runs the requested action and returns its result.
Check the artifact
Run tests, inspect the scene or verify the research against its sources.
Hosted access now, weights to follow
Moonshot released K3.1 Pro through its products and API on October 23. It says the full weights will be released by November 4, 2026. Teams planning to host the model should check the eventual license and deployment requirements.
K3.1 Pro always reasons. Its low, high and max effort settings let you choose how much work it spends before answering. Start with a bounded task and leave space for reasoning and the final answer in the output budget.
Try a coding task
Create a MiniRouter API key, set MINIROUTER_API_KEY in your environment, and run a prompt whose result you can test.
curl https://api.minirouter.sh/v1/chat/completions \
-H "Authorization: Bearer $MINIROUTER_API_KEY" \
-H 'Content-Type: application/json' \
--data '{
"model": "moonshotai/kimi-k3.1-pro",
"reasoning_effort": "low",
"max_tokens": 8192,
"messages": [{
"role": "user",
"content": "Write a Python CSV parser that handles quoted commas and escaped quotes. Include tests for both cases."
}]
}'Keep exploring
Compare published results in the model rankings and coding benchmarks, or read more model releases.
For MiniRouter product updates, visit the changelog.
Sources
- Read at Moonshot AI ↗
Moonshot Kimi K3.1 Pro announcement
Published Oct 23, 2026
- Read at Moonshot AI ↗
Kimi K3.1 Pro API guide
Published Oct 23, 2026
- Read at Vercel ↗
Vercel Kimi K3.1 Pro catalog
Published Oct 23, 2026
Editorial reference revised 2026-10-23. Catalog status comes from the generated routing snapshot.