JobBench: leaderboard
JobBench is a text benchmark for agentic workplace tasks across 130 tasks in 35 occupations, with workflows built from heterogeneous reference files. It evaluates models on expert-prioritized delegation work, not a replacement story. The pinned metric, Overall (%), measures average task success against fact-anchored binary criteria. Source: BenchmarkList’s attributed result table.
Metric: Overall (%). Source: benchmarklist.com. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark Open (xHigh) | 61.8 | #42 (Muse Spark Open) |
| 2 | GLM-5.3 (Max) | 61.4 | #44 (GLM-5.3) |
| 3 | Kimi K3 (Max) | 54.3 | #35 (Kimi K3) |
| 4 | Inkling (xHigh) | 31.1 | #144 (Inkling) |
| 5 | Nemotron 3 Ultra | 27.2 | #245 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=jobbench · How It Works · Data refreshed daily, snapshot 2026-11-05.