TestsBenchmarks · Agentic
OSWorld 2.0 offline set
Internet-free subset of OSWorld 2.0 (v2026.08.08 release), partial reward, as reported by OpenAI.
Measured by independent testersIndependent Reported by the makerVendor-reported
| Thinking levelSetting | |||||||
|---|---|---|---|---|---|---|---|
| 1 | GPT-6 Astra | OpenAI | 73.5% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 29 Sep 2026 |
| 2 | GPT-6.1 Sol | OpenAI | 71.4% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 29 Sep 2026 |
| 3 | Claude Opus 5 | Anthropic | 70.2% | Maxeffort=max | Independent testIndependentOpenAI (competitor result in OpenAI launch post) ↗ | — | 22 Sep 2026 |
| 4 | GPT-5.6 Sol | OpenAI | 66.2% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 22 Sep 2026 |
| 5 | GPT-6 Sol | OpenAI | 64.4% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 29 Sep 2026 |
| 6 | GPT-5.6 Luna | OpenAI | 52.7% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 22 Sep 2026 |
| 7 | GPT-6 Luna | OpenAI | 52.7% | Maxreasoning effort=max | Maker's own figureVendor-reportedOpenAI ↗ | — | 22 Sep 2026 |