Kimi K2.5, the latest open-source multimodal AI model from Moonshot AI, is making waves across the industry. Trained on 15 trillion tokens, it processes text, images, and video with remarkable speed and accuracy. In head-to-head testing, Kimi K2.5 outperforms GPT-5.2, Gemini 3 Pro, and Claude Opus on multiple benchmarks—particularly in coding and video reasoning. It beats Gemini 3 Pro on SWE-Bench Verified and demonstrates stronger multilingual coding abilities than GPT-5.2. On the VideoMMMU benchmark for video reasoning, Kimi K2.5 also takes the lead.
Moonshot AI has additionally launched Kimi Code, an open-source coding agent that competes with Claude Code and Gemini CLI. It supports terminal workflows, images, and video inputs, broadening the tool's appeal for developers.
With these advances, Kimi K2.5 is positioned as a top contender in the 2026 AI landscape, pushing the boundaries of what open-source models can achieve.