Quick start
ollama run kimi-k2-thinkingAvailable sizes
| Tag | Size | Quantization | Context | Min RAM |
|---|
Run with
Claude Code
ollama launch claude --model kimi-k2-thinking:cloudCodex
ollama launch codex --model kimi-k2-thinking:cloudOpenCode
ollama launch opencode --model kimi-k2-thinking:cloudOpenClaw
ollama launch openclaw --model kimi-k2-thinking:cloudStrengths & Limitations
Strengths
- Complex reasoning
- Open-source availability
- Strong thinking capabilities
Benchmarks
| Benchmark | Score | Unit |
|---|---|---|
| SWE-Bench Verified | 47.1 | % |
| AIME 2025 | — | — |
| GPQA-Diamond | 84.5 | — |
| MMLU-Pro | 84.6 | — |
| MMLU-Redux | 94.4 | — |
| SWE-bench Verified w/ tools | 71.3 | % |
| SWE-bench Multilingual w/ tools | 61.1 | % |
| SWE-bench w/ tools | 41.9 | % |
Related models
llama3.1Language
Llama 3.1 is a new state-of-the-art model from Meta available in 8B, 70B and 405B parameter sizes.
110.5M pullsllama3.2Language
Meta's Llama 3.2 goes small with 1B and 3B models.
58.0M pullsmistralLanguage
The 7B model released by Mistral AI, updated to version 0.3.
25.6M pullsqwen2.5Language
Qwen2.5 models are pretrained on Alibaba's latest large-scale dataset, encompassing up to 18 trillion tokens. The model supports up to 128K tokens and has multilingual support.
22.0M pulls