Kimi K2 ThinkingPricing
- Input Tokens: $0.548 /M tokens
- Output Tokens: $2.192 /M tokens
- Cache Read: $0 /M tokens
Input Modalities
- Text
Context length
- 262K tokens
Max output
- 262K tokens
Capabilities
- Thinking
- Streaming
- Tool calling
- Web search
- URL context
- Code interpreter
- Computer use
- File search
- Memory tool
- Structured outputs
- Citations
- Prompt caching
- Background mode
- Server-side sessions
Providers
Performance for kimi-k2-thinking
Uptime is the percentage of requests that succeeded over the past 72 hours. AIHubMix continuously monitors every provider and automatically retries with the next-best provider when one returns an error or responds too slowly; Latency is total round-trip time (lower is better); Throughput is how fast the model writes (tokens per second, higher is better).
Try this model
Frequently asked questions
More models from Moonshot AI
See all Moonshot AI models →Kimi-K2-0905 is a large-scale Mixture of Experts (MoE) language model developed by Moonshot AI, with a total of 1 trillion parameters and 32 billion active parameters per forward pass. It supports long-context inference of up to 256k tokens, an expansion from the previous 128k.
Kimi K2 is Moonshot AI's state-of-the-art Mixture-of-Experts (MoE) language model with 1 trillion total parameters and 32 billion activated parameters. Designed for agentic intelligence, it excels at tool use, coding, and autonomous problem-solving across diverse domains.
Kimi-K2 is a MoE architecture foundational model with extremely powerful coding and agent capabilities, featuring a total of 1 trillion parameters and activating 32 billion parameters. In benchmark performance tests across major categories such as general knowledge reasoning, programming, mathematics, and agents, the K2 model outperforms other mainstream open-source models. The Kimi-K2 model supports a context length of 128k tokens. It does not support visual capabilities.
The kimi-k2-turbo-preview model is a high-speed version of kimi-k2, with the same model parameters as kimi-k2, but the output speed has been increased from 10 tokens per second to 40 tokens per second.
The official billing is tiered based on input lengths of 8k, 32k, and 128k. This site does not support that billing structure and uses the middle 32k tier as the standard for charging. If you are price-sensitive, please avoid using it.
Kimi-Dev-72B is a new generation open-source programming large model that achieved a leading performance of 60.4% on SWE-bench Verified. Through large-scale reinforcement learning optimization, it can automatically fix code in real Docker environments, receiving rewards only when passing the complete test suite, thereby ensuring the correctness and robustness of solutions and aligning more closely with real software development standards.
