Agent Arena🏆Overall

Dynamic ranking of models on how well they orchestrate tools for real-world agentic tasks, based on signals like tool reliability, task completion, and steerability.

Oct 2, 2026

2,158,478 sessions

51 models

Pareto Frontier

Net Improvement x Cost/task from real Agent Mode tasks

CtrlScroll

Pareto Optimal Models

Net Improvement x Cost/task

Claude Fable 5.1 (Max)

Anthropic · Proprietary

+14.31%

$4.55/task

Claude Opus 5.5 (High)

Anthropic · Proprietary

+13.82%

$1.58/task

GPT 6.1 Sol (Max)

OpenAI · Proprietary

+11.23%

$0.56/task

Deepseek V4.1 Flash (Max)

DeepSeek · MIT

+4.02%

$0.10/task

MiMo V2.6 Pro

Xiaomi · MIT

+3.28%

$0.10/task

GPT 6 Luna (Max)

OpenAI · Proprietary

+1.35%

$0.07/task

MiMo V2.6 Flash

Xiaomi · MIT

-0.57%

$0.04/task

Mimo V2.5 Pro

Xiaomi · MIT

-7.52%

$0.04/task

Cost per Task

Median cost per task (last 14 days)

1Claude Fable 5.1 (Max)$4.55/task+14.31%

2Claude Opus 5.5 (High)$1.58/task+13.82%

3Claude Sonnet 5.5 (Max)$2.77/task+12.52%

4GPT 6 Astra (Max)$2.92/task+12.27%

5GPT 6.1 Sol (Max)$0.56/task+11.23%

6GPT 6 Sol (Max)$0.91/task+9.71%

7Claude Opus 5 (High)$2.79/task+8.67%

8Claude Fable 5 (High)$2.18/task+8.21%

9Claude Opus 5 (Max)$3.84/task+7.92%

10Gemini 4 Argon (High)$0.67/task+7.57%

11Claude Opus 4.8 (High)$1.37/task+6.64%

12GPT 5.6 Sol (xHigh)$1.05/task+6.45%

13Claude Sonnet 5 (High)$0.97/task+4.40%

14Kimi K3 (Max)$0.75/task+4.18%

15GPT 5.5 (xHigh)$1.12/task+4.15%