Arena.ai reports that GPT-6.2 Sol (Max) scored 1241 on Agent Arena, ranked 11th, and had a median cost of $0.90 per task. These are evaluation results published by Arena, not measurements independently reproduced locally.
Tasks this is suitable for assessing: Comparing the overall Agent Index, ranking, and cost per task in this leaderboard snapshot.
Tasks this should not be generalized to: Individual task performance. Quality and cost with other agents or toolchains. And model configurations not listed on this leaderboard.
Applicable model version: GPT-6.2 Sol (Max), model ID: gpt-6.2-sol.
Test environment or client: Arena Agent Arena. The specific runtime environment, toolset, and harness are not specified.
Reasoning tier and parameters: Max. Other reasoning parameters are not specified.
Arena.ai describes the Agent Arena Index as an overall agent metric built from millions of real-world tasks and trajectories contributed by a global community. It says the GPT-6.2 Astra and Sol results are based on 7.1K+ real agent sessions. The post does not explain how those 7.1K+ sessions were allocated across models, the task sampling and scoring details, how prices were calculated, or any confidence intervals.
GPT-6.2 Sol (Max): 1241 points, ranked 11th. Added to the cost Pareto frontier.
GPT-6.2 Sol (Max) scored the same 1241 points as GPT-6.1 Sol (Max). Arena says GPT-6.2 Sol's median cost per task fell 13%, from $1.04 to $0.90.
It scored 50 points below GPT-6.2 Astra (Max). Arena says Sol's median cost per task was 69% lower ($0.90 versus $2.94).
The rankings, scores, and costs above are reported by Arena in the original post. Rankings and costs use the platform's methodology and do not by themselves prove that Sol is cheaper for every task or delivers equivalent results.
| Model | Tier | Rank | Index score | Median cost per task reported in the post |
|---|---|---|---|---|
| GPT-6.2 Sol | Max | 11 | 1241 | $0.90 |
| GPT-6.1 Sol | Max | 12 | 1241 | $1.04 |
| GPT-6.2 Astra | Max | 6 | 1291 | $2.94 |
The narrow conclusion supported by these results is that, under Arena's published Agent Arena Index methodology, Sol (Max) matched GPT-6.1 Sol (Max) in score, Arena reported a lower cost, and Sol entered the cost Pareto frontier. The post does not provide enough information to independently recalculate the index or cost, and does not disclose the task distribution, sample breakdown, full tool configuration, or uncertainty.
The original post gives a cost of $0.90. At the time of collection, the live Agent leaderboard linked from the post showed a cost of $0.96 for GPT-6.2 Sol (Max), and said that cost was the median over the most recent 14 days. The two readings differ. This may reflect a leaderboard refresh or a change in the measurement window, but the page does not explain why. This article therefore retains the value from the original post. The two values should not be treated as readings from the same snapshot.
The results cannot be fully reproduced from the post: session-level samples, task and model sampling rules, tool and harness versions, price components, scoring formula, and statistical uncertainty are missing. The original post's English text was available at the time of collection. The results listed are data reported by Arena, and the date is recorded as shown in the post.
GPT-6.2 Sol