Kimi K3.1 Pro: Pricing, API Cost & Plans

Kimi K3.1 Pro is Moonshot’s new 1M-token model: limited free use, plans from $15 a month and API at $2 in, $10 out per 1M tokens. See costs and limits.

Moonshot AI released Kimi K3.1 Pro on October 23, 2026. It’s the first model in the new K3.1 series, a post-trained version of Kimi K3 that Moonshot calls its most capable 3T-class open-weight model so far, with a 1,048,576-token context window. You can use it in Kimi on the web, Kimi Work, Kimi Code and the Kimi API. We checked every price on Kimi’s K3.1 Pro pricing guide, the membership page and the API quickstart on October 25, 2026.

Is Kimi K3.1 Pro free?

Partly. The free Kimi plan (called Adagio) includes limited K3.1 Pro use on Kimi on the web and Kimi Work, so you can try it without paying. Kimi Code, the faster Highspeed version and the 1M-token extra-long chats need a paid plan.

The API isn’t free. K3.1 Pro is a flagship model on the Kimi API Platform, so it unlocks only after you top up your account (minimum $1), and then you pay per token.

Kimi K3.1 Pro Pricing (2026)

Kimi membership plans

Plan	Billed yearly	Billed monthly	What K3.1 Pro access you get
Free (Adagio)	$0	$0	Limited K3.1 Pro use on web and Kimi Work
Plus	$15/month ($180 a year)	$19/month	K3.1 Pro on web, Kimi Work and Kimi Code
Pro	$31/month ($372 a year)	$39/month	Adds K3.1 Pro Highspeed, Goal mode and Kimi Claw
Max	$79/month ($948 a year)	$99/month	5x agent credits and chats up to 1M tokens
Ultra	$159/month ($1,908 a year)	$199/month	10x agent credits and chats up to 1M tokens

Kimi K3.1 Pro API pricing

Model	Input	Output	Cache hit	Cache write (5 min / 1 h)
kimi-k3.1-pro	$2.00	$10.00	$0.20	$2.00 / $4.00
kimi-k3.1-pro-highspeed	$4.00	$20.00	$0.20	$4.00 / $8.00

Some quick maths. A request with 100,000 input tokens and 10,000 output tokens costs about $0.30 on standard K3.1 Pro and $0.60 on Highspeed. Filling the whole 1M-token window once costs $2 in input, or $0.20 if that context is already cached. Because K3.1 Pro always thinks before answering, reasoning tokens count as output, so check the reported output usage rather than the length of the final answer.

It’s also cheaper than the model it replaces. Kimi lists Kimi K3 at $3 input, $15 output and $0.30 for cache hits per 1M tokens, so standard K3.1 Pro is a third lower on each rate. If you want a sense of what a token is, our guide to tokens in AI explains it.

What’s new in Kimi K3.1 Pro

• 1M-token context: 1,048,576 tokens, with output up to the same limit (131,072 tokens by default).
• Highspeed version: 2-3x faster than standard K3.1 Pro, at twice the price for input, output and cache writes.
• Coding and 3D: Moonshot highlights full-project builds, game engines such as Unity, Unreal and Godot, and scripted modelling in Blender and CAD tools.
• Finance work: Moonshot reports gains over K3 of 7.2 points on its financial modelling test and 11.6 points on financial research.
• Fewer factual errors: in Moonshot’s own test on prompts where users had flagged mistakes, the claim-level hallucination rate fell from 23.35% for K3 to 6.85%.
• Open weights coming: Moonshot says the full model weights will be released by November 4, 2026.

Kimi K3.1 Pro vs Kimi K2

Our older Kimi K2 review covers the model that made Kimi known for long-document analysis. K3.1 Pro is two generations newer: it adds a 1M-token window, always-on reasoning with a choice of low, high or max effort, vision input and much stronger agent and coding skills. If you use Kimi in the browser, you don’t need to do anything; pick K3.1 Pro in the model menu. If you call the API, switch the model name to kimi-k3.1-pro.

Limitations

• Weights aren’t out yet: until November 4 you can only use it through Kimi’s apps and API.
• Thinking can’t be turned off: reasoning effort can be lowered, but every answer uses reasoning tokens.
• Fixed sampling: temperature, top_p and penalties are fixed, so leave them out of API requests.
• Image input: vision requests need base64 or ms:// file references; public image URLs aren’t supported.
• Web search: Moonshot says its API web search is being updated and isn’t recommended for production for now.
• Benchmarks pending: some benchmark scores on the launch page were still placeholders when we checked.

Kimi K3.1 Pro alternatives

• DeepSeek V4.1 Flash: a cheaper open-weight model for everyday coding and chat.
• MiniMax M3.1 Flash: another Chinese model with low API prices.
• Muse Glimmer: Meta’s 30B open agentic model that runs on your own hardware.

For a wider comparison, see our ranking of the best AI models.

Kimi K3.1 Pro gives you a 1M-token frontier model at $2 in and $10 out per 1M tokens, a third less than Kimi K3, with limited free use in the browser.

Try it free on Kimi first, use the API for long-context coding and research, and watch for the open weights in early November.