multimodal benchmark
OSWorld-Verified Leaderboard
OSWorld-Verified is a verified subset of OSWorld, a scalable real computer environment for multimodal agents supporting task setup, execution-based evaluation, and interactive learning across Ubuntu, Windows, and macOS.
Updated Aug 17, 2026
Models24
Model coverage24
MetricScore
EvidenceB
OSWorld-Verified Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAC | Score86.1% | Percentile100.0% | Participants24 | EvidenceC | Evaluated |
| Rank02 | ModelAN | Score85.0% | Percentile95.7% | Participants24 | EvidenceC | Evaluated |
| Rank03 | ModelAC | Score84.3% | Percentile91.3% | Participants24 | EvidenceC | Evaluated |
| Rank04 | ModelAN | Score83.4% | Percentile87.0% | Participants24 | EvidenceC | Evaluated |
| Rank05 | ModelGO | Score83.0% | Percentile82.6% | Participants24 | EvidenceC | Evaluated |
| Rank06 | ModelAN | Score81.2% | Percentile78.3% | Participants24 | EvidenceC | Evaluated |
| Rank07 | ModelME | Score80.8% | Percentile73.9% | Participants24 | EvidenceC | Evaluated |
| Rank08 | ModelAN | Score79.6% | Percentile69.6% | Participants24 | EvidenceC | Evaluated |
| Rank09 | ModelOP | Score78.7% | Percentile65.2% | Participants24 | EvidenceC | Evaluated |
| Rank10 | ModelGO | Score78.4% | Percentile60.9% | Participants24 | EvidenceC | Evaluated |
| Rank11 | ModelAN | Score78.0% | Percentile56.5% | Participants24 | EvidenceC | Evaluated |
| Rank12 | ModelOP | Score75.0% | Percentile52.2% | Participants24 | EvidenceC | Evaluated |
| Rank13 | ModelGO | Score74.0% | Percentile47.8% | Participants24 | EvidenceC | Evaluated |
| Rank14 | ModelAC | Score73.3% | Percentile43.5% | Participants24 | EvidenceC | Evaluated |
| Rank15 | ModelMA | Score73.1% | Percentile39.1% | Participants24 | EvidenceC | Evaluated |
| Rank16 | ModelOP | Score72.1% | Percentile34.8% | Participants24 | EvidenceC | Evaluated |
| Rank17 | ModelMI | Score70.1% | Percentile30.4% | Participants24 | EvidenceC | Evaluated |
| Rank18 | ModelME | Score65.9% | Percentile26.1% | Participants24 | EvidenceC | Evaluated |
| Rank19 | ModelOP | Score64.7% | Percentile21.7% | Participants24 | EvidenceC | Evaluated |
| Rank20 | ModelAC | Score62.5% | Percentile17.4% | Participants24 | EvidenceC | Evaluated |
| Rank21 | ModelAC | Score58.0% | Percentile13.0% | Participants24 | EvidenceC | Evaluated |
| Rank22 | ModelAC | Score56.2% | Percentile8.7% | Participants24 | EvidenceC | Evaluated |
| Rank23 | ModelAC | Score54.5% | Percentile4.3% | Participants24 | EvidenceC | Evaluated |
| Rank24 | ModelOP | Score39.0% | Percentile0.0% | Participants24 | EvidenceC | Evaluated |
OSWorld-Verified Highlights
The leading models and scores on this benchmark.
OSWorld-Verified Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for OSWorld-Verified
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
Ranking basisThis osworld-verified AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01ACQwen3.8 MaxAlibaba Cloud / Qwen Team
- Score
- 86.1%
- Price
- $2.0 input / $6.0 output per 1M tokens
Strengths
- Ranks #1 of 24 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures OSWorld-Verified, not total model capability
- 02ANClaude Fable 5Anthropic
- Score
- 85.0%
- Price
- $10 input / $50 output per 1M tokens
- Speed
- Up to 9.1 tok/s via Anthropic
Strengths
- Ranks #2 of 24 compared models
- 96th percentile on this benchmark
- C evidence result
Considerations
- This result measures OSWorld-Verified, not total model capability
- 03ACQwen3.8-27BAlibaba Cloud / Qwen Team
- Score
- 84.3%
Strengths
- Ranks #3 of 24 compared models
- 91th percentile on this benchmark
- C evidence result
Considerations
- This result measures OSWorld-Verified, not total model capability
- 04ANClaude Opus 4.8Anthropic
- Score
- 83.4%
- Price
- $5.0 input / $25 output per 1M tokens
- Speed
- Up to 46 tok/s via Anthropic
Strengths
- Ranks #4 of 24 compared models
- 87th percentile on this benchmark
- C evidence result
Considerations
- This result measures OSWorld-Verified, not total model capability
- 05GOGemini 3.6 FlashGoogle
- Score
- 83.0%
- Price
- $1.5 input / $7.5 output per 1M tokens
- Speed
- Up to 181 tok/s via Google
Strengths
- Ranks #5 of 24 compared models
- 83th percentile on this benchmark
- C evidence result
Considerations
- This result measures OSWorld-Verified, not total model capability
Selection summary
Best AI Models for OSWorld-Verified
Qwen3.8 Max currently leads OSWorld-Verified with 86.1%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is OSWorld-Verified?
What OSWorld-Verified measures and how its scores work.
OSWorld-Verified is a verified subset of OSWorld, a scalable real computer environment for multimodal agents supporting task setup, execution-based evaluation, and interactive learning across Ubuntu, Windows, and macOS.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- OSWorld-Verified
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- osworld-verified|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about OSWorld-Verified.
Which model scores highest on OSWorld-Verified?
Qwen3.8 Max is currently ranked first with 86.1%.
What does OSWorld-Verified measure?
OSWorld-Verified is a verified subset of OSWorld, a scalable real computer environment for multimodal agents supporting task setup, execution-based evaluation, and interactive learning across Ubuntu, Windows, and macOS.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
24 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.