Kimi K3.1 Pro: An Open-Weight Model Within a Point of Claude Opus 5.5 on Coding
Moonshot's new flagship scores 65.7% on Terminal-Bench 4.0 — 0.7 points behind Claude Opus 5.5 — and its weights are due by 4 November. What it is, what it costs, and the catch on cyber skills.
यह लेख हिंदी में पढ़ेंMoonshot AI launched Kimi K3.1 Pro on 23 October 2026, and on its own coding test it lands 0.7 points behind Claude Opus 5.5. It is a '3T-class' model with a 1 million token context, and the open weights are due by 4 November. You can use it today on kimi.com and through the Kimi API.
Where you can use it
On Kimi.com and the Kimi apps, in Kimi Code (Moonshot's coding agent), and through the API under the IDs kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Highspeed runs two to three times faster with the same abilities. You can't self-host it yet, because the full weights are only promised for 4 November.
One thing to plan for: K3.1 Pro only runs in thinking mode. Moonshot's docs say a request with thinking turned off is rejected, so reasoning tokens will be part of every bill.
Moonshot's benchmarks
| Benchmark | Kimi K3.1 Pro | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 65.7% | 66.4% | 58.2% |
| BenchCAD (with tools) | 92.7% | 96.2% | 95.9% |
| Kimi Code Bench 3.0, long-horizon | 61.2% (K3: 55.95%) | not given | not given |
These are Moonshot's numbers. Independent results are a bit more modest but still good: Vals.ai ranks it 8th of 27 models. Moonshot also says its hallucination rate on a hard internal stress test fell from 23.4% to 8.3%, and adds that this is a stress test, not a typical error rate.
What it costs
Moonshot's blog lists no prices. Third-party listings give about $2 per million input tokens and $10 per million output for K3.1 Pro, and roughly double that for Highspeed. Check Moonshot's pricing page before you budget. If that rate holds, it is the same headline price as Claude Sonnet 5.5 and half of Claude Opus 5.5.
The cyber catch
Moonshot deliberately weakened one skill for the open release. It removed cyber-security data from training, and its internal hacking benchmark fell from 90.5% to 81.0%. Moonshot says the open model's cyber ability stays behind frontier models that have been public for at least three months. Most people won't notice. If you do security research, you should know.
Is it worth trying?
Yes if you want close-to-frontier coding with the option to self-host later, which helps with data-residency rules. Not yet if you need it on your own servers today. Compare it with DeepSeek V4.1 Flash, which is MIT-licensed and far cheaper, and with Claude Sonnet 5.5 on your own tasks before you switch.
Sources
- Moonshot: Kimi K3.1 Pro tech blog
- Kimi API: K3.1 Pro quickstart
- Vals.ai: Kimi K3.1 Pro (independent ranking)
Frequently asked questions
Is Kimi K3.1 Pro open source?
Its weights are due to be released by 4 November 2026. Until then it is available only through Kimi's apps and API. Moonshot's blog does not state the licence.
How good is Kimi K3.1 Pro at coding?
On Moonshot's own Terminal-Bench 4.0 result it scores 65.7%, against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra. Vals.ai independently ranks it 8th of 27 models.
How much does the Kimi K3.1 Pro API cost?
Moonshot's blog does not list it. Third-party listings give $2 per million input tokens and $10 per million output, with Highspeed at roughly double. Check Moonshot's pricing page for the current rate.
Save this summary as an image or share it.
AICreatorHub Team
The AICreatorHub editorial team is a group of hands-on AI practitioners, writers and developers based in India. We test AI tools and models ourselves, track official releases from OpenAI, Anthropic, Google, Meta and xAI, and translate them into simple, India-first guides in English and Hindi. Every article is written for real Indian use cases — pricing in rupees, free-tier tips and practical, tested steps — so you get accurate, up-to-date and genuinely useful AI information.