Arena reports that GPT-6.2 Sol (Max) scored 1,757 in Code Arena: WebDev, ranking seventh and placing on the Pareto frontier. It scored 13 points higher than GPT-6.1 Sol (Max), with a 21% lower median cost per task.
Tasks this evaluation can inform: Preference-vote rankings in Arena Code Arena's WebDev category, and the median cost per task reported by that leaderboard.
Tasks this evaluation should not be extrapolated to: Other coding tasks, the quality of real-world project delivery, or other evaluation platforms. The post does not provide Sol's subcategory scores, vote count, or details of how costs were calculated.
Applicable model version: GPT-6.2 Sol (Max).
Test environment or client: Arena Code Arena: WebDev. The post describes the scores as real-world Arena scores but does not specify the client or runtime environment.
Reasoning level and parameters: Max. Other parameters were not disclosed.
The post reports model rankings and scores in Code Arena: WebDev, Pareto-frontier positions, and median cost per task, comparing GPT-6.2 Sol with GPT-6.1 Sol (Max) and GPT-6.2 Astra (Max). It does not disclose the vote count, number of tasks, cost basis, or specific statistical methods.
GPT-6.2 Sol (Max): ranked 7th, scored 1,757, and placed on the Pareto frontier.
Compared with GPT-6.1 Sol (Max): scored 13 points higher. Median cost per task fell from $1.84 to $1.45, a 21% decrease.
Compared with GPT-6.2 Astra (Max): scored 24 points lower. Median cost per task was 73% lower ($1.45 for Sol, $5.45 for Astra).
The original post did not publish GPT-6.2 Sol's WebDev subcategory ranking or score. The first-place Consumer Product result mentioned in the post belongs to GPT-6.2 Astra and cannot be treated as a Sol subcategory result.
| Model | WebDev rank | Score | Median cost per task |
|---|---|---|---|
| GPT-6.2 Sol (Max) | 7 | 1757 | $1.45 |
| GPT-6.1 Sol (Max) | Not listed | Not listed | $1.84 |
| GPT-6.2 Astra (Max) | 3 | 1781 | $5.45 |
The original post also reports that GPT-6.2 Sol scored 13 points higher than GPT-6.1 Sol while costing 21% less, and scored 24 points lower than GPT-6.2 Astra while costing 73% less.
This is a snapshot of the WebDev leaderboard published by Arena on 2026-11-18. It does not represent other tasks or evaluation environments. The post does not specify the vote count, task sample size, cost calculation, or subcategory results. Rankings and costs should be cited as published by the platform and should not be used to infer unpublished subcategory preferences.
The model rankings can be checked on the Arena Code Arena: WebDev leaderboard linked in the original post. The values provided in the post that can be checked are GPT-6.2 Sol (Max)'s 7th-place ranking, score of 1,757, and median cost per task of $1.45. The leaderboard is dynamic, so retain the collection date for this snapshot. The original post does not provide enough information to recalculate costs or reconstruct the voting sample.
GPT-6.2 Sol