Benchmarks · agentic
Agent Arena · Confirmed success
Agent Arena signal: how much a model raises the chance that a user confirms the task is done. A model's contribution to real agentic sessions, in percentage points.
45 models · 46 results · data as of 2026-10-29 · arena.ai ↗
Leaderboard
Opus 5.5 max
All results
46 rows| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Opus 5.5 | max | 22.3% ±3.1 |
| 2 | Sonnet 5.5 | max | 20.8% ±4.2 |
| 3 | Gemini 4 Carbon | high | 18.1% ±1.6 |
| 4 | Gemini 4 Argon | high | 12.8% ±1.5 |
| 5 | GPT-6 Astra | max | 12.6% ±3.5 |
| 6 | Fable 5.1 | max | 12.5% ±3.5 |
| 7 | Opus 5 | max | 11.5% ±2.3 |
| 8 | GPT-6.1 Sol | max | 11.4% ±1.9 |
| 9 | MiMo V2.6 Pro | high | 6.5% ±1.8 |
| 10 | MiMo V2.6 Flash | high | 6.5% ±2.4 |
| 11 | DeepSeek V4.1 Flash | max | 5.9% ±0.7 |
| 12 | Fable 5 | high | 5.6% ±2.1 |
| 13 | GPT-6 Sol | max | 5.4% ±3.4 |
| 14 | Grok 4.7 | xhigh | 4.5% ±3.1 |
| 15 | Hy4 | high | 4.4% ±1.5 |
| 16 | Kimi K3 | max | 4.3% ±1.2 |
| 17 | Muse Spark 1.3 | max | 3.7% ±1.8 |
| 18 | Qwen 3.8 Max | max | 3.5% ±1.4 |
| 19 | GLM 5.3 | max | 3.1% ±1.7 |
| 20 | GLM 5.2 | max | 2.9% ±1.6 |
| 21 | Haiku 5.5 | high | 2.6% ±2.4 |
| 22 | GLM 5.3 Flash | max | 2.3% ±1.6 |
| 23 | GPT-5.6 Sol | xhigh | 2.3% ±2.2 |
| 24 | Qwen 3.8 Flash | max | 1.2% ±1.7 |
| 25 | Gemini 3.8 Flash | high | 1.1% ±1.7 |
| 26 | Nemotron 3 Ultra 550B A55B | max | -22.1% ±7.8 |
| 27 | Sonnet 5 | high | -0.6% ±2.5 |
| 28 | DeepSeek V4 Pro | max | -2.9% ±3.8 |
| 29 | Gemini 3.1 Pro | high | -11.5% ±2.7 |
| 30 | Gemini 3.7 Flash | high | -2.9% ±1.9 |