Moonshot AI

Kimi-K2-Thinking

Generally availableOpen weights262K contextReasoning

Summary

Kimi-K2-Thinking is an open-weight language model from Moonshot AI, released on 6 Nov 2025. It accepts text and generates text, with a 262K-token context window and up to 262K output tokens per response. As of 18 Nov 2026, the lowest listed API price is $0.60 per million input tokens and $2.50 per million output tokens (Hugging Face Inference Providers), across 3 providers we track.

Kimi reasoning model for long-horizon research, planning, and tool use

Last verified 18 Nov 2026. Every figure on this page links to its source.

Specifications

DeveloperMoonshot AI
Release date6 Nov 2025 [models.dev]
StatusGenerally available [models.dev]
WeightsOpen weights [Hugging Face]
Licensemodified-mit [Hugging Face]
Parameters1T [Hugging Face]
Context window262K (262,144 tokens) [models.dev]
Max output262K tokens [models.dev]
Knowledge cutoffAug 2024 [models.dev]
InputText [models.dev]
OutputText [models.dev]
Reasoning modeYes [models.dev]
Tool callingYes [models.dev]
Structured outputYes [models.dev]
Hugging Facemoonshotai/Kimi-K2-Thinking [Hugging Face]

Kimi-K2-Thinking API pricing by provider

USD per million tokens. Sorted by input price.

ProviderInputOutputCached inputContextSource
Hugging Face Inference Providers
moonshotai/Kimi-K2-Thinking
$0.60$2.50$0.15262Kmodels.dev, 18 Nov 2026
Novita AI
moonshotai/kimi-k2-thinking
$0.60$2.50$0.15262Kmodels.dev, 18 Nov 2026
OpenRouter
moonshotai/kimi-k2-thinking
$0.60$2.50$0.15262KOpenRouter, 18 Nov 2026

Running Kimi-K2-Thinking locally

Estimated memory, from the parameter count and the architecture in config.json on Hugging Face. Includes about 0.8 GB of runtime overhead.

QuantizationWeightsTotal, 8K contextTotal, 32K context
Q4_K_M (4-bit)622 GB624 GB625 GB
Q8_0 (8-bit)1091 GB1092 GB1094 GB
FP16 / BF16 (16-bit)2053 GB2054 GB2056 GB

At 4-bit with an 8K conversation, Kimi-K2-Thinking needs about 624 GB, more than any single consumer graphics card; it needs a unified-memory machine, a multi-GPU setup or a split between graphics card and system RAM. Check your own hardware, or see Best computer for local AI on our sister site AIPCs.

Frequently asked questions

What is the context window of Kimi-K2-Thinking?

Kimi-K2-Thinking has a context window of 262,144 tokens (262K) and can generate up to 262,144 tokens in a single response, according to models.dev.

How much does the Kimi-K2-Thinking API cost?

Kimi-K2-Thinking is listed from $0.60 per million input tokens and $2.50 per million output tokens (Hugging Face Inference Providers). Prices last verified 18 Nov 2026.

Is Kimi-K2-Thinking open source?

Kimi-K2-Thinking is an open-weight model: its weights are published on Hugging Face as moonshotai/Kimi-K2-Thinking under the modified-mit license. Check the license terms before commercial use.

When was Kimi-K2-Thinking released?

Moonshot AI released Kimi-K2-Thinking on 6 Nov 2025, according to models.dev.

Which providers offer Kimi-K2-Thinking?

We track Kimi-K2-Thinking on 3 providers: Hugging Face Inference Providers, Novita AI, OpenRouter.

How much VRAM does Kimi-K2-Thinking need?

About 624 GB at 4-bit (Q4_K_M) with an 8K-token conversation, and about 1092 GB at 8-bit. That is more than any single consumer graphics card holds. These are estimates from the model's published parameter count and architecture.

Similar models from other labs

Comparable on output type, weights, release date, price and context window. Prices are the lowest listed, USD per million tokens.

ModelLabReleasedContextInput $/1MOutput $/1MWeights
Qwen3-235B-A22B-Thinking-2507Alibaba (Qwen)25 Jul 2025262K$0.30$3Open
GLM-4.6Zhipu AI (Z.ai)30 Sept 2025205K$0.50$2Open
Qwen3.5 397B-A17BAlibaba (Qwen)15 Feb 2026262K$0.45$3Open
GLM-4.7Zhipu AI (Z.ai)22 Dec 2025205K$0.40$1.75Open
Hermes-4-405BNousresearch26 Aug 2025131K$1$3Open
Magistral Small 1.2Mistral AI18 Sept 2025128K$0.50$1.50Open

Change history

  • Max output changed from 98K to 262K
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 262K to 98K
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 98K to 262K
  • Max output changed from 262K to 98K
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking
  • Max output changed from 98K to 262K
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 262K to 98K
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking
  • Max output changed from 98K to 262K
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 262K to 98K
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking
  • Max output changed from 98K to 262K
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 262K to 98K
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking
  • Name changed from Kimi K2 Thinking to Kimi-K2-Thinking
  • Max output changed from 98K to 262K
  • Max output changed from 262K to 98K
  • No longer listed on amazon-bedrock
  • Name changed from Kimi-K2-Thinking to Kimi K2 Thinking

Sources