🚨 GPT-5.6 Sol is really good at autonomous computer tasks

• Terminal-Bench 2.1: 91.9% • BrowseComp: 92.2% • OSWorld 2.0: 62.6% • ExploitBench: 73.5% • ARC-AGI-3: 7.78%, up from GPT-5.5’s 0.43%

The strongest results come from Sol’s Ultra multi-agent mode, which coordinates several agents across parallel work.

Have you tried GPT-5.6-Sol and if so what are your thoughts on it?