Moonshot AI released Kimi K2 Thinking in November 2025, an open-weights model with 1 trillion parameters that achieves state-of-the-art performance against closed models from OpenAI, Anthropic, and Google. The model utilizes a Mixture-of-Experts architecture activating 32B parameters per forward pass and native INT4 quantization to optimize efficiency.

  • Achieves 44.9% on Humanity's Last Exam (HLE) with tools, surpassing GPT-5 and Claude 4.5 Sonnet Thinking.
  • Scores 60.2% on BrowseComp and 71.3% on SWE-Bench Verified for agentic and coding tasks.
  • Features a 256K token context window capable of executing 200-300 sequential tool calls without human intervention.
  • Independent verification by Artificial Analysis confirmed a #1 ranking on the Tau2 Bench Telecom agentic benchmark at 93%.

The release demonstrates that open models can now compete with proprietary frontier systems, offering developers access to high-performance reasoning capabilities without licensing restrictions.