Google DeepMind's Gemini 4 Carbon (High) debuted at number 2 on Agent Arena's leaderboard on October 28, 2026, after over 21,900 sessions and a net improvement of 16.2%. The model runs at a median cost of $1.80 per task, which is 72% less than the top-ranked model.
Xiaomi's MiMo V2.6 Pro took the number 1 spot in Agent Arena's open-weight models category in mid-October 2026, posting a net improvement of 3.8% with costs as low as $0.11 per task. Open-weight models publish their trained parameters, allowing developers to download and run them on their own hardware.
Anthropic's Claude Opus 5.5 leads overall performance on Agent Arena with around a 20% net improvement as of early November 2026. Anthropic and OpenAI both show high raw performance, but the pressure appears on cost versus performance metrics. On the most expensive Agent Arena tasks, Google DeepMind and Xiaomi occupy frontier spots while OpenAI and Anthropic have slipped.
Agent Arena, run by arena.ai, judges AI models on their performance as agents. The platform launched in June 2026 and has gathered performance data from over 2 million real-world user sessions. The scoring relies on task success and user feedback rather than static benchmark scores.