Benchmarks · coding
VulcanBench Frontier v4
23 behavioural-reconstruction tasks: the model must repair a replacement implementation of a legacy program until hidden tests confirm it reproduces the program’s real drift from its written spec; the combined score weights functional correctness 50%, lint/complexity 8.5%, security 8.5% and judged Code quality 33%. Maintained by Morgan Linton. Score: Combined score across the 23 tasks, one column per model × reasoning-effort level: 100 × (0.50 × partial-credit functional hidden-test score + 0.085 × lint/complexity
13 models · 47 results · data as of 2026-11-06 · vulcanbench.com ↗
Leaderboard
Sonnet 5.5 max
All results
47 rows| # | Model | Effort | VulcanBench Frontier v4 | Harness |
|---|---|---|---|---|
| 1 | Sonnet 5.5 | max | 92.0% | Claude Code |
| 2 | Fable 5.1 | max | 91.8% | Claude Code |
| 3 | Opus 5.5 | medium | 90.9% | Claude Code |
| 4 | Fable 5.1 | xhigh | 90.8% | Claude Code |
| 5 | Opus 5.5 | xhigh | 90.7% | Claude Code |
| 6 | Opus 5.5 | max | 90.2% | Claude Code |
| 7 | Sonnet 5.5 | xhigh | 90.2% | Claude Code |
| 8 | Fable 5.1 | medium | 90.1% | Claude Code |
| 9 | Fable 5.1 | high | 90.1% | Claude Code |
| 10 | GPT-6.1 Terra | max | 90.0% | Codex |
| 11 | GPT-6 Astra | max | 89.3% | Codex |
| 12 | GPT-6 Astra | xhigh | 89.2% | Codex |
| 13 | Fable 5.1 | low | 89.1% | Claude Code |
| 14 | GPT-5.6 Terra | max | 89.1% | Codex |
| 15 | GPT-6.1 Sol | xhigh | 88.8% | Codex |
| 16 | GPT-6.1 Sol | max | 88.3% | Codex |
| 17 | GPT-6.1 Sol | high | 88.2% | Codex |
| 18 | GPT-6 Astra | high | 88.2% | Codex |
| 19 | Haiku 5.5 | high | 88.0% | Claude Code |
| 20 | GPT-6 Astra | medium | 87.7% | Codex |
| 21 | GPT-6 Astra | low | 87.4% | Codex |
| 22 | GPT-5.6 Sol | xhigh | 86.5% | Codex |
| 23 | GPT-5.6 Sol | high | 85.8% | Codex |
| 24 | GPT-5.6 Luna | max | 84.1% | Codex |
| 25 | Sonnet 5.5 | medium | 84.1% | Claude Code |
| 26 | GPT-6.1 Terra | xhigh | 83.7% | Codex |
| 27 | GPT-5.6 Terra | xhigh | 83.5% | Codex |
| 28 | GPT-6 Sol | high | 82.8% | Codex |
| 29 | GPT-5.6 Sol | medium | 82.6% | Codex |
| 30 | GPT-5.6 Luna | xhigh | 79.1% | Codex |