Model results · OpenAI

GPT-6.2 Sol

Suitev4
GPT-6.2 Sol across every effort level
In Codex across five effort levels: combined score 88.44 at Low, 88.94 at Medium, 88.86 at High, 89.01 at Extra-high and 88.74 at Max, nearly flat. Every task passed at every level, with every hidden test passed in all 115 runs; Code quality 70.31 to 71.91; 2.8 to 11.3 minutes per task. Every cell is judged on 23 runs.
November 19, 2026 · 23 tasks · 5 effort levels · 115 runs of GPT-6.2 Sol · VulcanBench Frontier v4 · Code quality protocol v3.31
→
Suitev4
Beside Claude Fable 5.2, Claude Opus 5.5 and GPT-6.1 Sol
GPT-6.2 Sol stays 3.3 to 5.1 points behind Claude Fable 5.2 and about 2 behind Claude Opus 5.5 from Medium up, at $0.23 to $0.55 a task against $1.70 to $10.88. It beats GPT-6.1 Sol by 2.2 to 2.5 points at Low and Medium and matches it above. On the Frontier v4 board its Extra-high level ranks 27th of 91 columns.
November 19, 2026 · comparison card · VulcanBench Frontier v4
→
Suitev4
Cost and tokens across effort levels
$0.23 to $0.55 per task at list API rates and 0.77M to 1.31M raw tokens per task across the five effort levels; the whole sweep prices at $45.38 for 115 runs, $0.39 per task.
November 19, 2026 · API-equivalent estimates at list rates, solver inference only · VulcanBench Frontier v4
→
Tracesv4
Redacted agent traces: GPT-6.2 Sol on Frontier v4
The trace repository for the sweep in the report above, published separately on GitHub. Structure only: every agent message, shell command, file change and tool call with its timing, and a closing summary per run, without the task content.
GitHub · vulcanbench-gpt62-sol-traces
→

How would GPT-6.2 Sol do on your stack?

Public results answer "which model is stronger on this suite." Whether GPT-6.2 Sol belongs in your routing table depends on your languages, your codebases, and your task mix, that gets measured, not guessed. Grading rules are in the methodology.