Strengths
- Cheaper than Kimi K3: $2 per million input and $10 per million output tokens versus $3 and $15.
- Moonshot reports that the share of false claims fell from 23.35% with K3 to 6.85%.
- Both context and maximum answer length can reach one million tokens, with answers capped at 131,072 tokens by default.
- It accepts text, images and video.
Weaknesses
- It is brand new. The official page still shows placeholder benchmark figures, and independent evaluations are scarce.
- Thinking cannot be switched off, which adds waiting time and paid output tokens.
- It does not accept images as public URLs, you have to upload them as files or embed them in the request.
- API data may be stored in Singapore, and Moonshot does not offer EU processing.
Best for
- agentic coding
- office documents and spreadsheets
- financial research
- very long outputs
Not ideal for
- data that must stay in the EU
- simple tasks where speed matters
How to use it in Europe
The model is on Kimi.com, in the Kimi mobile app and in the Kimi Work and Kimi Code tools. Through the international API at platform.kimi.ai you pay in USD, and the API unlocks after a top-up of at least $1. The API privacy policy says Moonshot may store data on servers in Singapore, and EU processing is not offered. Moonshot promises to release the weights by 4 November 2026. If it does, you will be able to run the model on your own servers, though at this size only on data-center hardware. Moonshot has not yet said which license the weights will use.
What is coming
Confirmed by vendor Moonshot AI launched Kimi K3.1 Pro in its API on 23 October 2026 and promised to publish the full weights by 4 November. Companies could then run it on their own servers. When: by 4 November 2026 (announced by Moonshot) →
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Claim-level hallucination rate (vendor-reported, lower is better) | 6.85% (Kimi K3: 23.35%) | platform.kimi.ai ↗ (vendor-reported), October 2026 |
| Terminal-Bench 4.0 (vendor-reported, via press) | 65.7% | letsdatascience.com ↗ (vendor-reported), October 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- Kimi K3: Moonshot AI's open model that topped the open-weight rankings in summer 2026, strong at web browsing and computer use.
Similar models from other vendors
- Claude Opus 5.5 (Anthropic): Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- DeepSeek V4 Pro (DeepSeek): DeepSeek's largest open-weight model for demanding text tasks at a low price. It does not understand images.
- Gemini 4 Carbon (Google): Google's most capable model, announced on 28 October 2026. So far only selected cybersecurity experts have access.
- Gemini 4 Argon (Google): Google's most capable publicly available model, out in October 2026. Built for demanding analytical work where accuracy matters and made-up answers are costly.
- Muse Wonder (Meta): Meta's strongest model for long-running agents and coding, still released as a preview.
- Mistral Large 4 (Mistral AI): Mistral's largest model, still in preview. Built for demanding tasks, long documents and security analysis with data kept in the EU.