Benchmarks · agentic
Agent Arena
Arena.ai Agent Arena: causal 'net improvement' a model contributes to real-world agentic sessions (task success, praise vs complaint, steerability, bash recovery, tool hallucination), estimated from millions of live Agent Mode sessions. Percent with 95% CI; not an Elo.
45 models · 46 results · data as of 2026-10-29 · arena.ai ↗
Leaderboard
Opus 5.5 max
All results
46 rows| # | Model | Effort | Agent Arena |
|---|---|---|---|
| 1 | Opus 5.5 | max | 20.9% ±2.2 |
| 2 | Gemini 4 Carbon | high | 16.0% ±1.0 |
| 3 | Sonnet 5.5 | max | 13.8% ±2.7 |
| 4 | Fable 5.1 | max | 13.4% ±1.9 |
| 5 | GPT-6 Astra | max | 13.0% ±2.0 |
| 6 | GPT-6.1 Sol | max | 12.5% ±1.1 |
| 7 | Gemini 4 Argon | high | 10.7% ±0.8 |
| 8 | Fable 5 | high | 8.0% ±1.1 |
| 9 | GPT-6 Sol | max | 8.0% ±1.6 |
| 10 | Opus 5 | max | 7.6% ±1.2 |
| 11 | GPT-5.6 Sol | xhigh | 4.8% ±1.0 |
| 12 | Sonnet 5 | high | 2.9% ±1.2 |
| 13 | MiMo V2.6 Pro | high | 2.8% ±0.9 |
| 14 | Gemini 3.8 Flash | high | 2.6% ±0.8 |
| 15 | Muse Spark 1.3 | max | 2.6% ±0.8 |
| 16 | Grok 4.7 | xhigh | 2.4% ±1.5 |
| 17 | GPT-6 Luna | max | 2.2% ±0.4 |
| 18 | DeepSeek V4.1 Flash | max | 1.9% ±0.3 |
| 19 | Haiku 5.5 | high | 1.4% ±1.1 |
| 20 | GLM 5.2 | max | 1.3% ±0.8 |
| 21 | Hy4 | high | 1.2% ±0.7 |
| 22 | Kimi K3 | max | 1.1% ±0.6 |
| 23 | MiMo V2.6 Flash | high | 1.0% ±1.1 |
| 24 | GLM 5.3 | max | 0.9% ±0.8 |
| 25 | Grok 4.6 | xhigh | 0.5% ±1.1 |
| 26 | Qwen 3.8 Max | max | 0.4% ±0.6 |
| 27 | Nemotron 3 Ultra 550B A55B | max | -15.8% ±3.3 |
| 28 | DeepSeek V4 Pro | max | -0.3% ±1.7 |
| 29 | Gemini 3.1 Pro | high | -4.5% ±1.3 |
| 30 | Gemini 3.7 Flash | high | -1.3% ±0.9 |