Inference
EN
VulcanBench contributors

VulcanBench Frontier v4

An open coding benchmark of 23 hard behavioural-reconstruction tasks, run at every reasoning-effort level and scored on correctness and judged code quality.

ModelsModels with a result.17
Top resultBest result on this benchmark.—
Top-3 spreadPoints from first to third.—
YearYear of release.2026

Result and price

15 of 17 results

Results

17 results
#ModelResultThe score from the source.
—Grok 4.793.2%
—Claude Haiku 5.592.2%
—Claude Sonnet 5.592.0%
—Claude Fable 5.191.8%
—Claude Opus 5.591.1%
—GPT-6.1 Terra90.0%
—GPT-6 Astra89.3%
—GPT-5.6 Terra89.0%
—GPT-6.1 Sol88.8%
—DeepSeek V4.1 Flash 101788.7%
—Muse Wonder Full88.4%
—GPT-5.6 Sol87.2%
—GPT-6 Sol86.8%
—GPT-5.6 Luna84.2%
—GPT-5.578.5%
—GPT-6 Luna71.9%
—Beam 1.0 501B-A23B64.7%