What Is Kimi K3.1 Pro?

TL;DR

• Kimi K3.1 Pro is Moonshot AI’s current flagship model and the first model in its K3.1 series.
• It is a 3-trillion-class model with a 1,048,576-token context window, native vision input, structured output, and tool calls.
• Moonshot highlights long-horizon coding, agentic work, 3D creation, finance research, and factual accuracy as major upgrades.
• API list price is ¥13 per 1M input tokens and ¥68 per 1M output tokens; the Highspeed variant costs ¥26 / ¥136.

What Kimi K3.1 Pro is

Kimi K3.1 Pro is a frontier reasoning model exposed through the Kimi Open Platform as model ID kimi-k3.1-pro. It is designed for long-running engineering, knowledge, multimodal, and agent workflows rather than only short chat replies.

The model always uses thinking mode and supports reasoning_effort values low, high, and max. Developers can combine that control with vision input, JSON Schema outputs, tool calls, dynamic tool definitions, partial completion, streaming, and automatic prefix caching.

Moonshot also publishes kimi-k3.1-pro-highspeed, a same-capability variant advertised as producing output approximately two to three times faster for double the standard token price.

Kimi K3.1 Pro specifications

What improves over Kimi K3

Kimi K3 is Moonshot’s previous 2.8-trillion-parameter flagship. K3.1 Pro moves to a 3-trillion-class model and increases reinforcement-learning compute across hundreds of downstream domains.

The clearest published gains are in long engineering and knowledge workflows: Moonshot reports 61.2% on Kimi Code Bench 3.0 Long-horizon versus 55.95% for K3, and 65.7% on Terminal-Bench 4.0.

For factual robustness, Moonshot’s stress test reports a single-fact error rate of 6.85% for K3.1 Pro versus 23.35% for K3. That is a vendor evaluation, so production teams should still benchmark their own documents and workflows.

Where K3.1 Pro is most useful

• Long coding agents that must inspect a repository, run terminal commands, fix failures, and preserve task state.
• Financial research and spreadsheet workflows requiring cross-document checking, formulas, and numerical reasoning.
• Product and creative engineering with screenshots, video, UI references, 3D scenes, or game systems.
• Support and research assistants that repeatedly load the same long knowledge base and can benefit from prefix caching.

Practical limitations

• Web search is listed as being updated and is not currently recommended for production flows in Moonshot’s guide.
• Vision input does not accept a public image URL directly; use base64 or an ms:// file ID.
• Thinking output consumes tokens, so unbounded reasoning can raise cost and latency.
• Official benchmarks are useful directional evidence, not a substitute for your own evaluation suite.