Kimi K2.6: Highest Open-Weight Model on SWE-Bench Verified at 80.2% — Within 0.6 Points of Opus 4.6, Sustained 4,000+ Tool Calls Over 13-Hour Session

Kimi·medium signal

MoonshotAI's Kimi K2.6 (1T parameters, released April 20) is the highest-ranked open-weights model with an Intelligence Index score of 54, scoring 80.2% on SWE-Bench Verified (just under Opus 4.6's 80.8%) and leading with 58.6% on SWE-Bench Pro. It sustained 4,000+ tool calls over a 13-hour uninterrupted session in published benchmarks, demonstrating production-grade agentic endurance. Available via Ollama for local deployment. For builders evaluating open-source alternatives for coding agents or self-hosted pipelines, K2.6 represents the first open model within striking distance of frontier proprietary models on real-world coding tasks.