# OSWorld-Verified — AI model rankings

Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Higher is better.

28 tracked models have a published OSWorld-Verified score. Higher is better. Scores come from published lab reports and benchmark sources; recorded sources appear beside the scores.

## Ranking

| Rank | Model | Developer | Score | Source | Released |
| --- | --- | --- | --- | --- | --- |
| 1 | Qwen3.8-Max | Qwen | 86.1% | Lab | Aug 3 2026 |
| 2 | Claude Fable 5 | Anthropic | 85% | Lab | Jun 9 2026 |
| 2 | Claude Mythos 5 | Anthropic | 85% | [BenchLM](https://benchlm.ai), retrieved 2026-08-24 | Jun 9 2026 |
| 4 | Qwen3.8-27B | Qwen | 84.3% | [BenchLM](https://benchlm.ai), retrieved 2026-08-24 | Aug 14 2026 |
| 5 | Claude Opus 4.8 | Anthropic | 83.4% | Lab | May 28 2026 |
| 6 | Gemini 3.6 Flash | Google | 83% | Lab | Jul 21 2026 |
| 7 | Claude Sonnet 5 | Anthropic | 81.2% | Lab | Jun 30 2026 |
| 8 | Muse Spark 1.1 | Meta | 80.8% | Lab | Jul 9 2026 |
| 9 | GPT-5.5 | OpenAI | 78.7% | Lab | Apr 23 2026 |
| 10 | Gemini 3.5 Flash | Google | 78.4% | Lab | May 19 2026 |
| 11 | Claude Opus 4.7 | Anthropic | 78% | Lab | Apr 16 2026 |
| 12 | Gemini 3.1 Pro | Google | 76.2% | Lab | Feb 19 2026 |
| 13 | GPT-5.4 | OpenAI | 75% | Lab | Mar 5 2026 |
| 14 | Gemini 3.5 Flash-Lite | Google | 74% | Lab | Jul 21 2026 |
| 15 | Qwen3.7-Plus | Qwen | 73.3% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Jun 1 2026 |
| 16 | Kimi K2.6 | Moonshot AI | 73.1% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Apr 21 2026 |
| 17 | Claude Opus 4.6 | Anthropic | 72.7% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Feb 5 2026 |
| 18 | Claude Sonnet 4.6 | Anthropic | 72.5% | Lab | Feb 17 2026 |
| 19 | GPT-5.4 mini | OpenAI | 72.1% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Mar 17 2026 |
| 20 | Claude Opus 4.5 | Anthropic | 66.3% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Nov 24 2025 |
| 21 | Muse Glimmer | Meta | 65.9% | Lab | Aug 10 2026 |
| 22 | Gemini 3.0 Flash | Google | 65.1% | Lab | Dec 17 2025 |
| 23 | GPT-5.3-Codex | OpenAI | 64.7% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Feb 5 2026 |
| 24 | Qwen3.5 | Qwen | 62.2% | Lab | Feb 16 2026 |
| 25 | Claude Sonnet 4.5 | Anthropic | 61.4% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Sep 29 2025 |
| 26 | Muse Spark | Meta | 53.3% | Lab | Apr 8 2026 |
| 27 | GPT-5.2 | OpenAI | 47.3% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Dec 11 2025 |
| 28 | GPT-5.4 nano | OpenAI | 39% | [BenchLM](https://benchlm.ai), retrieved 2026-08-31 | Mar 17 2026 |


---

Canonical page: https://aireleasetracker.com/benchmark/osworld-verified
Full dataset: https://aireleasetracker.com/llms-full.txt · JSON: https://aireleasetracker.com/models.json
Source: AI Release Tracker (https://aireleasetracker.com). Most benchmark scores come from lab launch material; gathered results identify the leaderboard that published them.