OSWorld-Verified leaderboard
As of November 8, 2026, Kimi K3 (Moonshot) leads OSWorld-Verified at 84.8%, ahead of Gemini 3.6 Flash at 83% and MiMo-V2.6-Pro at 82%. 8 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is OSWorld-Verified? OSWorld-Verified tests computer-use agents on real tasks in a live Ubuntu desktop: web and desktop apps, files and multi-app workflows, controlled through screenshots, mouse and keyboard. Made by XLANG Lab at the University of Hong Kong with Salesforce Research, CMU and the University of Waterloo, July 2025.
Top models on OSWorld-Verified, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | Kimi K3 Moonshot (Kimi) | max effort | $3 / $15 | 1M | Vendor-reported Moonshot AI |
| 2 | Gemini 3.6 Flash Google | as reported | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 3 | MiMo-V2.6-Pro Xiaomi MiMo | as reported | $0.435 / $0.87 | 1M | Vendor-reported Xiaomi |
| 4 | MiMo-V2.6-Flash Xiaomi MiMo | as reported | $0.14 / $0.28 | 1M | Vendor-reported Xiaomi |
| 5 | Gemini 3.5 Flash Deprecated Google | as reported | $1.5 / $9 | 1M | Vendor-reported Google DeepMind |
| 6 | Gemini 3.1 Pro Preview Preview Google · listed as “Gemini 3.1 Pro” | as reported | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 7 | Gemini 3.5 Flash-Lite Google | as reported | $0.3 / $2.5 | 1M | Vendor-reported Google DeepMind |
| 8 | Gemini 3.1 Flash-Lite Deprecated Google | as reported | $0.25 / $1.5 | 1M | Vendor-reported Google DeepMind |
OSWorld-Verified: score against price
Can you trust OSWorld-Verified scores?
No independent leaderboard publishes OSWorld-Verified 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.
- Verified is OSWorld after the maintainers fixed 300+ reported issues, so its scores are not comparable with older OSWorld numbers.
- The step budget changes the score a lot; a number without its step limit is hard to compare.
- OSWorld 2, a separate and much harder benchmark, has its own page here.
How OSWorld-Verified works
The task
The agent gets a natural-language instruction and a virtual desktop; it sees screenshots and issues mouse and keyboard actions until the task is done.
The score
Success rate: each task has its own script that checks the final state of the machine, and the score is the share of tasks passed.
Test set
369 tasks, as published by the authors.
Made by
XLANG Lab at the University of Hong Kong with Salesforce Research, CMU and the University of Waterloo, first published July 2025.
Official sources
- os-world.github.io
- Paper: arxiv.org/abs/2404.07972
- Definition read on 2026-11-08
Sources and freshness
8 published scores from 4 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 Google DeepMind
- Vendor-reported Google DeepMind
- Vendor-reported Xiaomi
- Vendor-reported Moonshot AI
OSWorld-Verified FAQ
Answers derived from the 8 published scores on this page and the current price index.