VulcanBench tests GPT-6.2 Sol on Frontier v4
VulcanBench tested OpenAI's GPT-6.2 Sol model across all five effort levels using Codex CLI, finding that the combined score remained nearly flat between 88.44 and 89.01. The benchmark results indicate that higher effort levels buy execution time and token usage rather than significant score improvements, with functional correctness remaining at 100% across all runs. Open article →