Kimi K3.1 Pro: Moonshot's new flagship is live
A new flagship has landed in NeuralSpace: Kimi K3.1 Pro by Moonshot AI. It is a reasoning model with a context window of up to 1 million tokens and a post-trained successor to Kimi K3, built for long-horizon work: deep analysis, large-scale code and multi-step reasoning. It is already available in the chat, the Code section and the public API. The model's own page is /chat/kimi-k3.1-pro.
What Kimi K3.1 Pro can do
- Deep reasoning. The model thinks before it answers, and the effort is adjustable: in the chat you get off, low, high and max modes — from quick replies to the most thorough analysis.
- 1M-token context. Very long documents, codebases and threads fit into a single request — the model keeps the thread across the whole volume.
- Web search. One checkbox in the chat: ask for fresh information and the model grounds its answer in search results.
- Coding. In the Code section the model works as an agent: it creates a project, writes and edits files and runs long iterations — exactly the long-horizon coding it was built for.
- Multimodality. Per the supplier catalog, the model accepts not only text but also images and video as input.
How K3.1 Pro differs from Kimi K3
K3.1 Pro is a direct, post-trained successor to the Kimi K3 flagship: the same open-weight line, with improved reasoning and long-horizon performance. Both models live side by side on the site: Kimi K3 and the new Kimi K3.1 Pro. If you want maximum depth and the longest context — go for K3.1 Pro.
Pricing
You pay for the tokens actually used, after the answer. The tariff is dynamic — it follows the supplier's live catalog and the exchange rate: at publication time that is $2.00 per 1M input tokens and $10.00 per 1M output tokens by the supplier catalog. The minimum balance threshold in the chat is 7 tokens. Current prices are on the API page.
How to try it
- Open the chat and pick Kimi K3.1 Pro in the model list — the model's own page is /chat/kimi-k3.1-pro.
- Ask your question; enable web search with a checkbox and pick a reasoning effort — from fast "off" to max.
- For projects, open the Code section — there the model works as an agent and edits files step by step.
- For your own apps, use the public API with a key: the model list and prices are on the API page. The model is also available via MCP.
When to pick K3.1 Pro
Kimi K3.1 Pro is the pick for heavy work: parsing long documents and reports, refactoring and architecting large codebases, multi-step analysis where reasoning depth matters. For quick everyday answers, a Flash-line model is cheaper — for example DeepSeek V4.1 Flash 1017.
FAQ
What is Kimi K3.1 Pro? Moonshot AI's flagship reasoning model and a post-trained successor to Kimi K3, with a 1M-token context. Available in the chat, the Code section, the public API and MCP.
What does "reasoning" mean? The model thinks before answering; in the chat the effort is adjustable: off, low, high, max.
Is there web search? Yes: one "Web search" checkbox in the chat.
How much does it cost? You pay for the tokens actually used, after the answer; the chat's minimum balance threshold is 7 tokens. The tariff is dynamic — current prices are on the API page.