Kimi-K3.1-Pro is the first model in Moonshot AI's new K3.1 series and its capable 3T-class open-weight model to date. It is a post-trained version of Kimi K3 with substantially larger post-training compute, featuring a 1-million-token context window and native visual understanding (text and image input, text output). Through large-scale joint reinforcement learning across hundreds of downstream domains, it delivers significant gains in coding, 3D creation, financial research, and other long-horizon agentic tasks: scoring 65.7% on Terminal-Bench 4.0, achieving a 64.4 on OfficeQA Pro v2, and reducing claim-level hallucination rate from 23.4% (K3) to 8.3%. Its overall intelligence approaches closed-source frontier models.
Inference Service Provider
The inference service provider for kimi-k3.1-pro is QwenCloud.
Model Capabilities
The following capabilities are the same across all regions and deployment scopes.
| Capability | Support |
|---|---|
| Input modalities | Text, image |
| Output modality | Text |
| Video input | Not supported |
| Model playground | Supported |
| Function Calling | Supported |
| Structured output | Supported |
| Implicit caching | Supported |
| Web search | Supported |
| Prefix continuation | Not supported |
| Batch inference | Not supported |
| Fine-tuning | Not supported |
Context Limits
| Parameter | Value (tokens) |
|---|---|
| Context length | 1048576 |
| Maximum input length | 1048576 |
| Maximum output length | 1048576 |
| Maximum input length (thinking mode) | 1048576 |
| Maximum output length (thinking mode) | 1048576 |
| Maximum chain-of-thought length | 1048576 |
Pricing
This page shows only the original pricing for model API calls, excluding limited-time promotions.
- Singapore
- China North 2 (Beijing)
- Germany (Frankfurt)
- China (Hong Kong)
- Japan (Tokyo)
- US (Virginia)
Scope: International
| Billable item | Price (USD) | Unit |
|---|---|---|
| Input | $2 | Per million tokens |
| Output | $10 | Per million tokens |
| Input (cache hit) | $0.2 | Per million tokens |
Rate Limits
- Singapore
- China North 2 (Beijing)
- Germany (Frankfurt)
- China (Hong Kong)
- Japan (Tokyo)
- US (Virginia)
Scope: InternationalThis region uses dynamic rate limiting. The TPM limit is tiered by your monthly QwenCloud spend, with three tiers of 1,200,000, 5,000,000, and 10,000,000. For the tier rules, see Dynamic rate limiting.
How to Call
- Model usage guide: Kimi.
- API references: OpenAI-compatible Chat Completions, OpenAI-compatible Responses, DashScope, and Anthropic-compatible Messages.