Benchmark
OSWorld-Verified
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.
- Models and systems measured
- 40
- Scores recorded
- 81
- Sources
- 2
- IQ weight
- 2.20
Score distribution
- 73.2%
- Mean
- 76.2%
- Median
- 10.6%
- Std dev
- 58%
- 10th percentile
- 83.4%
- 90th percentile
- 39%
- Min
This benchmark has a weight of 2.20 for chat models, measured over 38 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model or system | Score |
|---|---|---|
| 51 | Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 73.1% |
| 52 | Claude Opus 4.6Opus 4.6 System Card setting· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.7% |
| 53 | GPT-5.6 LunaGemini 3.6 Flash Model Card · July 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.6% |
| 54 | Claude Sonnet 4.61080p · 100 steps · pass@1 · 5 runs· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.5% |
| 55 | Claude Sonnet 4.6Gemini 3.5 Flash Model Card · May 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.5% |
| 56 | GPT-5.4 mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 72.1% |
| 57 | GPT-5.4 minixhigh · published setting· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 72.1% |
| 58 | MiniMax M3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 70.1% |
| 59 | MiniMax M3nogdrive 361 · 1920×1080 · 200 steps· Reported by MiniMax Research· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 70.1% |
| 60 | Claude Opus 4.5Opus 4.6 System Card setting· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 66.3% |
| 61 | Claude Opus 4.51080p · max 100 steps · pass@1 · avg@5 · 64K thinking · 200K context· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 66.3% |
| 62 | Muse Glimmer 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 65.9% |
| 63 | Muse Glimmer 30Bhigh reasoning· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.9% |
| 64 | Gemini 3 FlashGemini 3.5 Flash Model Card · May 2026· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.1% |
| 65 | GPT-5.3 Codex· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 64.7% |
| 66 | GPT-5.3 Codexxhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.7% |
| 67 | Qwen3.6 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 62.5% |
| 68 | Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.5% |
| 69 | Qwen3.5 397B A17BQwen3.5 release/model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62.2% |
| 70 | Claude Sonnet 4.51080p · max 100 steps · pass@1· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.4% |
| 71 | Claude Sonnet 4.5Opus 4.6 System Card setting· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 61.4% |
| 72 | Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 58% |
| 73 | Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 56.2% |
| 74 | Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 54.5% |
| 75 | NVIDIA Nemotron 3.5 Super VL 120B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 51.1% |
| 76 | Claude Haiku 4.51080p · max 100 steps · pass@1· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 50.7% |
| 77 | GPT 5.2OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 47.3% |
| 78 | GPT 5.4 Nano· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 39% |
| 79 | GPT 5.4 Nanoxhigh · published setting· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 39% |
Showing 51–79 of 79 results.
Not ranked
The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.
| Model or system | Score |
|---|---|
| Muse Spark (version unspecified)Meta Model API · highest effort · version unspecified · 361-task split excluding Google Drive · GUI-only · 1920×1080 · 200 steps · pass@1· Reported by Meta, Meta Superintelligence Labs· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 53.3% |
| Qwen3.8-2.4T-A95Blifecycle-page core-benchmark claim· Reported by Alibaba Cloud Model Studio· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 86.1% |
Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats