# OSWorld 2.0 Leaderboard

> Last updated SEP 03, 2026

Tracking reported computer-use agent performance on OSWorld 2.0, combining the official benchmark leaderboard with additional results found by Yutori Scout.

Official results come from the OSWorld 2.0 leaderboard. The official leaderboard has not been updated since Jun 25, 2026. We’ve set up a [Scout](<https://yutori.com/scouting>) to monitor for additional results and catch results it has missed.

## 150 steps

| Model | Organization | Reasoning | Tool setting | Partial score |
| --- | --- | --- | --- | --- |
| GPT-5.5 | OpenAI | XHigh | Batch Tool | 46.7% |
| [Navigator n2](<https://yutori.com/blog/introducing-n2>)<sup>1</sup> | Yutori | Medium | Batched Tool | 45.5% |
| Claude Opus 4.7 | Anthropic | Max | Standard | 20.3% |
| Claude Sonnet 4.6 | Anthropic | Max | Standard | 20.0% |
| MiniMax M3 | MiniMax | Enabled | Standard | 8.2% |
| Kimi 2.6 | Moonshot AI | Enabled | Standard | 7.1% |

<sup>1</sup> Navigator n2’s 500-step score is reported in the [launch post](<https://yutori.com/blog/introducing-n2>). Its 150- and 300-step scores, reasoning level, and tool setting are first-party Yutori data supplied for this leaderboard.

## 300 steps

| Model | Organization | Reasoning | Tool setting | Partial score |
| --- | --- | --- | --- | --- |
| [Navigator n2](<https://yutori.com/blog/introducing-n2>)<sup>1</sup> | Yutori | Medium | Batched Tool | 62.0% |
| GPT-5.5 | OpenAI | XHigh | Batch Tool | 49.5% |
| Claude Opus 4.7 | Anthropic | Max | Standard | 39.8% |
| Claude Sonnet 4.6 | Anthropic | Max | Standard | 35.8% |
| MiniMax M3 | MiniMax | Enabled | Standard | 16.6% |
| Qwen 3.7-Plus | Alibaba | Thinking | Standard | 16.6% |
| Kimi 2.6 | Moonshot AI | Enabled | Standard | 14.4% |

<sup>1</sup> Navigator n2’s 500-step score is reported in the [launch post](<https://yutori.com/blog/introducing-n2>). Its 150- and 300-step scores, reasoning level, and tool setting are first-party Yutori data supplied for this leaderboard.

## 500 steps

| Model | Organization | Reasoning | Tool setting | Partial score |
| --- | --- | --- | --- | --- |
| Claude Fable 5.1 | Anthropic | Max | – | 77.9% |
| GPT-6 Astra | OpenAI | Max | – | 72.6% |
| Claude Opus 5 | Anthropic | – | – | 70.6% |
| Muse Spark 1.3 | Meta | Max | – | 66.9% |
| Claude Fable 5 | Anthropic | – | – | 66.1% |
| [Navigator n2](<https://yutori.com/blog/introducing-n2>)<sup>1</sup> | Yutori | Medium | Batched Tool | 65.2% |
| GPT-5.6 Sol | OpenAI | Max | Batch Tool | 62.7% |
| Gemini 3.8 Flash | Google | – | Batched Tool | 59.0% |
| Kimi K3 | Moonshot AI | Max | – | 58.3% |
| Claude Opus 4.8 | Anthropic | Max | Batched Tool | 54.8% |
| Gemini 3.7 Flash | Google | – | Batched Tool | 50.6% |
| GPT-5.6 Terra | OpenAI | Max | – | 50.2% |
| GPT-5.5 | OpenAI | XHigh | Batch Tool | 49.5% |
| Claude Opus 4.7 | Anthropic | Max | Standard | 49.1% |
| Muse Spark 1.1 | Meta | XHigh | Batched Tool | 47.3% |
| GPT-5.6 Luna | OpenAI | – | – | 45.6% |
| Claude Sonnet 4.6 | Anthropic | Max | Standard | 41.5% |
| Gemini 3.6 Flash | Google | – | – | 33.8% |
| Gemini 3.1 Pro | Google | High | Batched Tool | 30.6%<sup>2</sup> |
| MiniMax M3 | MiniMax | Enabled | Standard | 22.3% |
| Kimi 2.6 | Moonshot AI | Enabled | Standard | 22.1% |
| Qwen 3.7-Plus | Alibaba | Thinking | Standard | 21.5% |

<sup>1</sup> Navigator n2’s 500-step score is reported in the [launch post](<https://yutori.com/blog/introducing-n2>). Its 150- and 300-step scores, reasoning level, and tool setting are first-party Yutori data supplied for this leaderboard.

<sup>2</sup> [Gemini 3.1 Pro result reported by Alibaba in the Qwen3.8 release post.](<https://qwen.ai/blog?id=qwen3.8>)

## About the benchmark

OSWorld 2.0 evaluates long-horizon computer-use workflows across self-hosted web environments and professional desktop applications. Each task contains multiple executable checkpoints, supporting partial-credit scoring for meaningful progress on workflows that were not completed end to end.

## Sources

- [Paper (arXiv 2606.29537)](<https://arxiv.org/abs/2606.29537>)
- [Code (GitHub · XLANG)](<https://github.com/xlang-ai/OSWorld-V2>)
- [Dataset (Hugging Face)](<https://huggingface.co/datasets/xlangai/osworld_v2_tasks>)
- [Official leaderboard](<https://osworld-v2.xlang.ai/#leaderboard>)
- [Official results (JSON)](<https://osworld-v2.xlang.ai/static/data/leaderboard/official-results.json>)
- [Navigator n2 launch post](<https://yutori.com/blog/introducing-n2>)

Official source revision (SHA-256): `a48d24c886c62a64336dc827c5b3cb46c61822de38522785965f804ba76f4aaf`