JobBench leaderboard
As of November 8, 2026, MiMo-V2.6-Pro (Xiaomi MiMo) leads JobBench at 62%, ahead of MiMo-V2.6-Flash at 61.2% and Kimi K3 at 54.3%. 5 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is JobBench? JobBench tests AI agents on the work that experts in 35 occupations say they most want to hand off, packaged as workspaces of real reference files. Made by Yuetai Li and 23 co-authors (University of Washington, UC Santa Barbara, Stanford, CMU and others), May 2026.
Top models on JobBench, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | MiMo-V2.6-Pro Xiaomi MiMo | as reported | $0.435 / $0.87 | 1M | Vendor-reported Xiaomi |
| 2 | MiMo-V2.6-Flash Xiaomi MiMo | as reported | $0.14 / $0.28 | 1M | Vendor-reported Xiaomi |
| 3 | Kimi K3 Moonshot (Kimi) | max effort | $3 / $15 | 1M | Vendor-reported Moonshot AI |
| 4 | Qwen3.8 Max Alibaba (Qwen) | as reported | $2 / $6 | 1M | Vendor-reported Alibaba (Qwen) |
| 5 | Qwen3.7 Max Alibaba (Qwen) | as reported | $2.5 / $7.5 | 1M | Vendor-reported Alibaba (Qwen) |
JobBench: score against price
Can you trust JobBench scores?
No independent leaderboard publishes JobBench results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- Grading is done by a model judge, and the project changed judge models after a retirement, so older and newer scores can differ slightly.
- At publication the best of 36 models reached 45.9%.
How JobBench works
The task
Each task is a workspace of mixed reference files from real professional work; the agent has to work through the clutter and produce the deliverable the duty calls for.
The score
A chain of fact-anchored yes-or-no rubrics, about 36 criteria per task on average, judged by a model and reported as a weighted percentage.
Test set
130 tasks across 35 occupations, as published by the authors.
Made by
Yuetai Li and 23 co-authors (University of Washington, UC Santa Barbara, Stanford, CMU and others), first published May 2026.
Official sources
- job-bench.github.io
- Paper: arxiv.org/abs/2605.26329
- Definition read on 2026-11-08
Sources and freshness
5 published scores from 3 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported Alibaba (Qwen)
- Vendor-reported Xiaomi
- Vendor-reported Moonshot AI
JobBench FAQ
Answers derived from the 5 published scores on this page and the current price index.