OmniaBench: leaderboard
OmniaBench evaluates general-purpose agents on 644 executable, verifiable challenging tasks spanning software, research, and everyday workflows; scored by overall Pass@1.
Metric: Overall Pass@1 (%). Source: scuuy.github.io. Saturation forecast: Around January 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5 | 58.54 | #57 |
| 2 | GPT-5.6 Sol | 57.14 | #19 |
| 3 | GLM-5.2 | 56.83 | #73 |
| 4 | GPT-5.5 | 56.52 | #31 |
| 5 | DeepSeek V4 Pro | 54.5 | #102 |
| 6 | Claude Opus 4.7 | 54.19 | #49 |
| 7 | Kimi K2.6 (Thinking) | 52.33 | #104 (Kimi K2.6) |
| 8 | Qwen 3.7 Max | 49.69 | #75 |
| 9 | DeepSeek V4 Flash | 48.6 | #137 |
| 10 | Gemini 3.5 Flash | 45.65 | #60 |
| 11 | GLM-5 (Thinking) | 43.32 | #141 (GLM-5) |
| 12 | Seed 2.1 Pro | 42.7 | #67 |
| 13 | Qwen 3.6 27B | 42.08 | #170 |
| 14 | Gemini 3.1 Pro (Preview) | 41.3 | #58 |
| 15 | Qwen 3.5 397B A17B | 41.15 | #146 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=omniabench · How It Works · Data refreshed daily, snapshot 2026-10-20.