Everything about Kimi K2 Thinking

What you need to know

Moonshot AI released Kimi K2 Thinking on November 6, 2025. It has a reasoning mode: it can "think" before answering, which helps with multi-step problems but uses more tokens, and therefore more money. Its weights are published: you can download it and run it on your own servers, or use one of the 18 API providers that list it on models.dev. That makes it 11 months old as of October 22, 2026.

We file it under open-weight models (68 models in this category on the site).

Measured performance

According to results published by Epoch AI (checked on October 22, 2026), Kimi K2 Thinking has a capability index of 69.6 out of 100, computed on 3 benchmarks from our basket.

With only 3 benchmarks reported, it does not enter the public ranking, which requires at least 4 results, so its index should be read as partial.

• GPQA Diamond (PhD-level science questions): 84.2% — the best tracked model, Gemini 4 Argon, reaches 96.4%
• OTIS Mock AIME (competition mathematics problems): 83.1% — the best tracked model, GPT-5.5 Pro, reaches 100%
• Terminal-Bench (agent tasks in a terminal): 35.7% — the best tracked model, GPT-5.5, reaches 84.7%

How much does Kimi K2 Thinking cost?

Kimi K2 Thinking costs $0.60 per million input tokens and $2.5 per million output tokens (median of 18 providers, checked on October 22, 2026).

To put it in perspective: reading 20,000 tokens (about thirty pages) and writing 2,000 costs roughly $0.017; a thousand of those tasks, $17.

That is 2.5 times the median output price of open-weight models ($1).

To estimate your own budget, use the agent cost calculator.

What it can do

Its context window reaches 262,144 tokens, roughly 393 pages of text at 500 words per page: enough for a long report or several contracts at once. A single answer can run to 262,144 tokens.

It accepts as input: text. Its knowledge stops at June 2025: beyond that, it needs web search or your documents.

It can call tools (functions, APIs), which makes it a possible foundation for an AI agent.

The weights are published on Hugging Face (moonshotai/Kimi-K2-Thinking). See also our list of open-source models.

When to choose it

Given its category, Kimi K2 Thinking is mainly aimed at projects where you want control over your data or hosting, or want to adapt the model to your field.

Before adopting it, test it on your own examples: benchmarks measure standard tasks, not your job.

What could replace it?

In the same category, the models closest to Kimi K2 Thinking are:

• GLM-5.2 (Zhipu AI (Z.ai), index 70.5)
• GLM-4.7 (Zhipu AI (Z.ai), index 68.4)
• DeepSeek V4 Flash 0731 (DeepSeek, index 68.2)
• DeepSeek V4 Pro 0813 (DeepSeek, index 72.2)