GPT 6.1 Sol vs. Claude Sonnet 5.5: A Three-Task Agent Test
实测丨GPT 6.1 Sol 硬刚 Sonnet 5.5：一个太老实，另一个精得很
- Who
- 雷峰网
- Quote
我们把两款模型放进同一个工作台，通过同一个渠道商调用，要求模型连续完成 3D 密室开发、最新政策调查和诺奖化学科普三项任务。
What happened
Leiphone tested GPT 6.1 Sol and Claude Sonnet 5.5 on 3D escape-room development, policy research, and chemistry education using Claude Code CLI and models accessed through OpenRouter. Sol completed the game faster and produced more extensive source archives and verification records, while Sonnet finished both research tasks much faster and spent $10.9 across the three tasks, compared with Sol’s $38.5. The results varied by task, with tool use and verification shaping completion time and cost.
Read next
See the full storyLeiphone Compares GPT 6.1 Sol and Claude Sonnet 5.5 on Three Agent TasksThe story so far
Leiphone tested GPT 6.1 Sol and Claude Sonnet 5.5 on three agent tasks using Claude Code CLI and models accessed through OpenRouter. Sol completed a 3D escape-room game faster and produced larger source archives and more verification records; Sonnet completed the policy research and chemistry education tasks much faster. Across all three tasks, Sonnet spent $10.9, compared with $38.5 for Sol. The results varied by task. Leiphone said tool use and verification affected both completion time and cost.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-11-03