GPT‑6.1 Sol lands fifth in Arena's Agent Arena at $0.56 a task

Arena said OpenAI's GPT-6.1 Sol (Max) entered its Agent Arena leaderboard in fifth place, up 11.23% net, in a post on X on Friday. Arena put the model's median cost at $0.56 per task and said it reshaped the leaderboard's Pareto frontier of cost against score.

By Arena's figures, GPT-6.1 Sol scores within two percentage points of GPT-6 Sol and GPT-6 Astra. It costs 39% less than GPT-6 Sol while scoring 1.52 points higher, and 81% less than GPT-6 Astra while landing within 1.04 points of it.

Arena also compared the model with the leader. It said GPT-6.1 Sol costs 88% less than Claude Fable 5.1 (Max), ranked first, while landing within 3.08 points. The post was cut off in the data we reviewed, so Arena's comparisons with other models, including a 65% lower cost against a Claude model, are not reported here in full.

Arena's Agent Arena ranks models on real agentic sessions, and its cost figures are its own medians, not prices OpenAI has published. We have not seen the underlying session counts for this entry, and rankings on such boards can shift as more votes arrive. OpenAI has not commented in the material reviewed.