WorkBench 2026 - Side Effects: leaderboard
Metric: Side-effect rate (%): share of tasks in which the agent took a wrong, harmful action (for example emailing the wrong person) on the corrected 2026 release of WorkBench: 690 templated workplace tasks over a sandbox of five databases (calendar, email, website analytics, CRM, project board) and 26 tools, graded by comparing the final sandbox state with the ground truth; one native tool-calling agent loop for every model, up to 20 steps, temperature 0 where allowed; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5 | 1.9 |
| 2 | Claude Opus 4.8 | 2.5 |
| 3 | Gemini 3.1 Pro (Preview) | 2.9 |
| 4 | Gemini 3.5 Flash | 3 |
| 5 | GPT-5.5 | 3.9 |
| 6 | Kimi K2.6 | 6.8 |
| 7 | Claude Sonnet 4.6 | 8.8 |
| 8 | DeepSeek V4 Pro | 12.8 |
| 9 | GPT-5 | 13 |
| 10 | GPT-4o | 15.1 |
| 11 | Claude Haiku 4.5 | 16.7 |
| 12 | GPT-5.4 | 16.8 |
| 13 | GLM-4.6 | 17 |
| 14 | O3 | 17.5 |
| 15 | GPT-5.1 | 18.1 |
Interactive version: theaggregate.ai/benchmark?slug=workbench-2026-side-effects · How It Works · Data refreshed daily, snapshot 2026-09-29.