OrchBench: leaderboard
Deterministic simulation of multi-agent orchestration plans on 10-task DAGs. The unit-interval score averages output quality, scheduling efficiency and token efficiency; larger DAG panels are separate experiments.
Metric: Score (0-1, 10-task DAGs). Source: arxiv.org. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 0.81 | #31 |
| 2 | Gemini 3.1 Pro (Preview) | 0.8 | #58 |
| 3 | GLM-5.1 | 0.8 | #97 |
| 4 | DeepSeek V4 Pro | 0.78 | #102 |
| 5 | Claude Opus 4.8 | 0.78 | #37 |
| 6 | DeepSeek V4 Flash | 0.77 | #137 |
| 7 | Kimi K2.6 | 0.77 | #104 |
| 8 | Qwen 3.6 35B A3B | 0.77 | #226 |
| 9 | Doubao Seed 2.0 Mini | 0.76 | #184 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=orchbench · How It Works · Data refreshed daily, snapshot 2026-10-20.