GPT 6.1 Sol and Claude Sonnet 5.5 split on speed, cost, and verification in agent tasks
AILeiphone tested GPT 6.1 Sol and Claude Sonnet 5.5 through Claude Code on three tasks: a 3D escape room, a policy investigation, and a chemistry explainer. Both models cost the same per token on OpenRouter, yet Sol's total bill was $38.5 against $10.9 for Sonnet 5.5, about 3.5 times higher.
Source details
Leiphone · leiphone.comPublished · added here
Original title: 实测丨GPT 6.1 Sol 硬刚 Sonnet 5.5：一个太老实，另一个精得很