OS-Critic Bench
Models judge whether a proposed action advances a task using the screenshot and action history. The grader compares each judgment with a human label.
Shanghai AI Laboratory compares 13 model configurations on 438 mobile, 152 desktop, and 148 web examples. It reports accuracy and F1.
- Made by
- Shanghai AI Laboratory and collaborators
- Released
- December 2025
- Kind
- Finding things on screen
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Written by crowdworkers
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
GPT-5 leads with 68.2% among models released by September 2025 GPT-5 leads with 68.2% among models released by September 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 GPT-5 68.2%
- 2 OS-Oracle-7B 68%
- 3 Gemini-2.5-Pro (snapshot unspecified) 67.6%
- 4 Claude Sonnet 4.5 66.9%
- 5 OS-Oracle-7B-SFT 63.1%
- 6 GPT-4o (snapshot unspecified) 62.2%
- 7 Qwen3-VL-8B 61%
- 8 GUI-Critic-R1 59.5%
- 9
Qwen2.5-VL-7B 58.3%
- 10
UI-TARS-1.5 57.5%
- 11 ScaleCUA-32B 51%
- 12 SE-WSM-7B 48.9%
- 13 ScaleCUA-7B 45.9%
Source: The OS-Critic Bench paper
- 1 GPT-5 68.2%
- 2 Claude Sonnet 4.5 66.9%
- 3
Qwen2.5-VL-7B 58.3%
- 4
UI-TARS-1.5 57.5%
Source: The OS-Critic Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 68.2% | Aug 2025 | Official leaderboard | |
| 2 | OS-Oracle-7B | 68% | Official leaderboard | ||
| 3 | Gemini-2.5-Pro (snapshot unspecified) | 67.6% | Official leaderboard | ||
| 4 | Claude Sonnet 4.5 | 66.9% | Sep 2025 | Official leaderboard | |
| 5 | OS-Oracle-7B-SFT | 63.1% | Official leaderboard | ||
| 6 | GPT-4o (snapshot unspecified) | 62.2% | Official leaderboard | ||
| 7 | Qwen3-VL-8B | 61% | Official leaderboard | ||
| 8 | GUI-Critic-R1 | 59.5% | Official leaderboard | ||
| 9 | | 58.3% | Jan 2025 | Official leaderboard | |
| 10 | | 57.5% | Apr 2025 | Official leaderboard | |
| 11 | ScaleCUA-32B | 51% | Official leaderboard | ||
| 12 | SE-WSM-7B | 48.9% | Official leaderboard | ||
| 13 | ScaleCUA-7B | 45.9% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 68.2% | Aug 2025 | Official leaderboard | |
| 2 | Claude Sonnet 4.5 | 66.9% | Sep 2025 | Official leaderboard | |
| 3 | | 58.3% | Jan 2025 | Official leaderboard | |
| 4 | | 57.5% | Apr 2025 | Official leaderboard |
Rankings by part
We read 6 parts from the benchmark’s paper, each ranking models on one part of the test, such as Mobile accuracy, Mobile F1 and Desktop accuracy. Claude Sonnet 4.5 leads 3 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Mobile accuracy
4 models we track have a score here.
- 1 GPT-5 69.2%
- 2 Claude Sonnet 4.5 65.3%
- 3
Qwen2.5-VL-7B 58.5%
- 4
UI-TARS-1.5 56.2%
Mobile F1
4 models we track have a score here.
- 1 GPT-5 69.5%
- 2
UI-TARS-1.5 ↑2 66.4%
- 3
Qwen2.5-VL-7B 65.8%
- 4 Claude Sonnet 4.5 ↓2 63.5%
Desktop accuracy
4 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 74.3%
- 2 GPT-5 ↓1 73.7%
- 3
Qwen2.5-VL-7B 61.2%
- 4
UI-TARS-1.5 58.5%
Desktop F1
4 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 77.5%
- 2 GPT-5 ↓1 74.4%
- 3
UI-TARS-1.5 ↑1 69.6%
- 4
Qwen2.5-VL-7B ↓1 69.4%
Web accuracy
4 models we track have a score here.
- 1 Claude Sonnet 4.5 ↑1 64.2%
- 2
UI-TARS-1.5 ↑2 60.1%
- 3 GPT-5 ↓2 59.5%
- 4
Qwen2.5-VL-7B ↓1 54.7%
Web F1
4 models we track have a score here.
- 1
UI-TARS-1.5 ↑3 70%
- 2 Claude Sonnet 4.5 64.9%
- 3
Qwen2.5-VL-7B 64.2%
- 4 GPT-5 ↓3 55.2%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The OS-Critic Bench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official OS-Critic Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Computer and browser use
- BrowseComp In use Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness. Claude Fable 5 93.3%
- ScreenSpot-Pro In use GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element. GPT-6 Astra 92.7%
- AndroidWorld In use Agents complete 116 tasks across 20 real Android apps in an emulator. Each task checks the device's system state to judge success. GPT-6 Astra 100%
- OSWorld 2.1 In use Agents complete 108 long-horizon workflows across 31 self-hosted websites and desktop apps, averaging 318 tool calls. Scored by binary completion and checkpoint partial credit. Claude Opus 5 44.3%
- WebArena-Verified In use Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces.
Muse Spark 1.1 69%
- OSWorld 2.1 (offline subset) In use Agents complete the 82 OSWorld 2.1 tasks on the official offline list, with no internet access. Scored by checkpoint partial credit and strict pass rate. Claude Fable 5.2 89.3%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (13 entries).
- Pass Models scored 4 models we track, plus 9 other entries on its leaderboard.
- Warning Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Pass Headroom The best is 68.2%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 4 rows for models we track and 9 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.