agentic Agentic · llm-stats.com
OSWorld
Real computer-use tasks across operating systems: browsers, office apps and file management.
3
Modelli testati
Claude Opus 4.6
Modello di punta
72.7%
Punteggio migliore
Anthropic
Provider
| # | Modello | OSWorld |
|---|---|---|
| 1 | Claude Opus 4.6 Anthropic | 72.7% |
| 2 | GPT-6 Astra OpenAI | 72.6% |
| 3 | Claude Sonnet 4.6 Anthropic | 72.5% |
Dati benchmark aggregati da llm-stats.com. Vedi Metodologia.