Moonshot AI has released Kimi K2, a new 128-billion parameter open-source model designed for complex coding tasks and agentic reasoning. This Mixture-of-Experts model, trained on 14.7 trillion tokens, rivals top closed-source models like Claude 3.5 Sonnet and GPT-4o on reasoning benchmarks. Kimi K2 is available under the Apache 2.0 license and shows strong performance in math and coding evaluations, with support for tool use and a long context window. AI
IMPACT Sets a new benchmark for open-source models in coding and agentic reasoning, potentially accelerating development in these areas.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=2 ai=1.0]
- Kimi K2.7 Code
- Moonshot AI
- Apache 2.0
- Claude 3.5 Sonnet
- DeepSeek V4
- GPT-4o
- Hugging Face
- Kimi K2
- Qwen 3
- vLLM
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →