# Kimi K2 Thinking API Pricing

As of October 10, 2026, Kimi K2 Thinking costs $0.60 per 1M input tokens and $2.50 per 1M output tokens from Azure AI Foundry, the lowest of 9 providers. It has a 262,144-token context window and returns up to 262,144 tokens per response.

- Page: https://models.litellm.ai/models/kimi-k2-thinking
- Lab: Moonshot AI
- LiteLLM model name: `azure_ai/kimi-k2-thinking`
- Context window: 262,144 tokens
- Max output: 262,144 tokens
- Added to LiteLLM: November 2025
- Deprecation date: March 29, 2026
- Prices updated: October 10, 2026

## Pricing by provider

Token prices in USD per 1M tokens.

| Provider | Model name | Input | Output | Cache read | Per image | Per second | Context |
|---|---|---|---|---|---|---|---|
| Azure AI Foundry | `azure_ai/kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| Amazon Bedrock | `bedrock/moonshotai.kimi-k2-thinking` | $0.73 | $3.03 | – | – | – | 256K |
| Amazon Bedrock | `moonshot.kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| Amazon Bedrock Mantle | `bedrock_mantle/moonshotai.kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| Baseten | `baseten/moonshotai/Kimi-K2-Thinking` | $0.60 | $2.50 | – | – | – | – |
| Crusoe | `crusoe/moonshotai/Kimi-K2-Thinking` | $2.50 | $2.50 | – | – | – | 256K |
| Fireworks AI | `fireworks_ai/accounts/fireworks/models/kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| GMI Cloud | `gmi/moonshotai/Kimi-K2-Thinking` | $0.80 | $1.20 | – | – | – | 256K |
| Google Vertex AI | `vertex_ai/moonshotai/kimi-k2-thinking-maas` | $0.60 | $2.50 | $0.06 | – | – | 256K |
| Novita AI | `novita/moonshotai/kimi-k2-thinking` | $0.60 | $2.50 | $0.15 | – | – | 256K |
| Amazon Bedrock (ap-northeast-1) | `bedrock/ap-northeast-1/moonshotai.kimi-k2-thinking` | $0.73 | $3.03 | – | – | – | 256K |
| Amazon Bedrock (ap-south-1) | `bedrock/ap-south-1/moonshotai.kimi-k2-thinking` | $0.71 | $2.94 | – | – | – | 256K |
| Amazon Bedrock (sa-east-1) | `bedrock/sa-east-1/moonshotai.kimi-k2-thinking` | $0.73 | $3.03 | – | – | – | 256K |
| Amazon Bedrock (us-east-1) | `bedrock/us-east-1/moonshotai.kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| Amazon Bedrock (us-east-2) | `bedrock/us-east-2/moonshotai.kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |
| Amazon Bedrock (us-west-2) | `bedrock/us-west-2/moonshotai.kimi-k2-thinking` | $0.60 | $2.50 | – | – | – | 256K |

## Cost examples

1,000 requests with 2,000 input and 500 output tokens each:

- Azure AI Foundry: $2.45
- Amazon Bedrock: $2.45
- Amazon Bedrock Mantle: $2.45
- Baseten: $2.45

100 long-document requests with 100,000 input and 2,000 output tokens each:

- Azure AI Foundry: $6.50
- Amazon Bedrock: $6.50
- Amazon Bedrock Mantle: $6.50
- Baseten: $6.50

## Features

Supported: Function calling, Parallel tool calls, Structured output, Reasoning, Web search.
Not supported: Image input, Video input, Prompt caching.

## Use it with LiteLLM

```python
from litellm import completion

response = completion(
    model="azure_ai/kimi-k2-thinking",
    messages=[{"role": "user", "content": "Hello!"}],
)
```

## FAQ

### How much does Kimi K2 Thinking cost?

Kimi K2 Thinking costs $0.60 per 1M input tokens and $2.50 per 1M output tokens from Azure AI Foundry, the lowest of 9 providers.

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

Kimi K2 Thinking has a 262,144-token context window and returns up to 262,144 tokens per response.

### Which providers offer Kimi K2 Thinking?

Kimi K2 Thinking is available from 9 providers through LiteLLM: Azure AI Foundry, Amazon Bedrock, Amazon Bedrock Mantle, Baseten, Crusoe, Fireworks AI, GMI Cloud, Google Vertex AI and Novita AI. The lowest price is on Azure AI Foundry.

### How do I call Kimi K2 Thinking with an OpenAI-compatible API?

Use the LiteLLM Python SDK with model="azure_ai/kimi-k2-thinking", or add the model to the LiteLLM proxy's config.yaml and send requests to /v1/chat/completions from any OpenAI client. LiteLLM tracks the cost of every request at the prices on this page.

### What features does Kimi K2 Thinking support?

Kimi K2 Thinking supports function calling, parallel tool calls, structured output, reasoning and web search. It does not support image input, video input and prompt caching.

### When did LiteLLM add Kimi K2 Thinking?

Kimi K2 Thinking was added to LiteLLM's model price file in November 2025.

### Is Kimi K2 Thinking being deprecated?

Azure AI Foundry deprecated azure_ai/kimi-k2-thinking on March 29, 2026.

Source: LiteLLM model price file, https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json