Kimi K2.5

Kimi K2.5, released by Moonshot AI in January 2026, is a natively multimodal model trained on 15 trillion mixed visual and text tokens, capable of understanding text, images, and video. It builds on the K2 architecture with significant improvements in agentic capabilities and coding performance.

K2.5 leads on multiple agentic benchmarks with 74.9% on BrowseComp (vs 59.2% for competitors), demonstrating exceptional web browsing and information synthesis. In coding, it outperforms Gemini 3 Pro on SWE-Bench Verified (76.8%) and beats both GPT-5.2 and Gemini 3 Pro on SWE-Bench Multilingual. On LiveCodeBench v6, it achieves 85.0% pass@1. It also delivers 76% lower operational costs compared to comparable models.

A key innovation is K2.5's agent swarm capabilities, enabling it to coordinate multiple specialized agents for parallel workflows. Within days of its release, Kimi K2.5 became the most-used AI model by token count, surpassing Claude Opus. Its combination of multimodal understanding, agentic performance, and aggressive pricing has made it a formidable competitor in the global AI landscape.