WorkflowArena - MATH - Resolved: leaderboard
WorkflowArena v1, 145 tasks across eight domains, native Resolved on MATH. Resolved requires all acceptance criteria; continuous outcome credits partial progress. Overall is the equal-weight domain mean. Codex 0.164.0 / Claude Code 2.1.280, exact native harness and effort retained. No transfer from the separate 32-task workflow-transfer cohort, no cross-harness or compute pooling.
Metric: Resolution Rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 25 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (WorkflowArena Run: harness=Claude Code; effort=max; protocol=145-task-v1) | 95.83 | |
| 2 | Claude Opus 5.5 (WorkflowArena Run: harness=Claude Code; effort=xhigh; protocol=145-task-v1) | 91.67 | |
| 3 | Claude Opus 5.5 (WorkflowArena Run: harness=Claude Code; effort=high; protocol=145-task-v1) | 87.5 | |
| 4 | GPT-6 Astra (WorkflowArena Run: harness=Codex; effort=xhigh; protocol=145-task-v1) | 83.33 | |
| 5 | GPT-6 Astra (WorkflowArena Run: harness=Codex; effort=high; protocol=145-task-v1) | 83.33 | |
| 6 | GPT-6.1 Sol (WorkflowArena Run: harness=Codex; effort=max; protocol=145-task-v1) | 83.33 | |
| 7 | GPT-6 Astra (WorkflowArena Run: harness=Codex; effort=max; protocol=145-task-v1) | 83.33 | |
| 8 | Claude Sonnet 5.5 (WorkflowArena Run: harness=Claude Code; effort=max; protocol=145-task-v1) | 79.17 | |
| 9 | GPT-6.1 Sol (WorkflowArena Run: harness=Codex; effort=xhigh; protocol=145-task-v1) | 79.17 | |
| 10 | GPT-6 Astra (WorkflowArena Run: harness=Codex; effort=medium; protocol=145-task-v1) | 79.17 |
Interactive version: theaggregate.ai/benchmark?slug=workflowarena-math-resolved · How It Works · Data refreshed daily, snapshot 2026-11-12.