OdysseyBench: leaderboard
OdysseyBench evaluates long-horizon office agents across Word, Excel, PDF, email, and calendar workflows. It includes OdysseyBench+ tasks from real-world use cases and OdysseyBench-Neo synthetic tasks that require multi-step reasoning over interaction history. (Source: benchmarklist.com, self-reported.)
Metric: Overall (Old) (self-reported). Source: benchmarklist.com. Saturation forecast: Around November 2026. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 | 56.19 | #124 |
| 2 | GPT-5 | 54 | #98 |
| 3 | GPT-5 Chat | 40.33 | #241 |
| 4 | DeepSeek R1 | 39.33 | #251 |
| 5 | GPT-4.1 | 35.67 | #248 |
| 6 | GPT-4o | 33.67 | #343 |
| 7 | Qwen 3 32B | 27 | #443 |
| 8 | O3 Mini | 23.75 | #272 |
| 9 | GPT-4o Mini | 19 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=odysseybench · How It Works · Data refreshed daily, snapshot 2026-10-20.