Photonic PCFBench - Refactor Syntax: leaderboard
PCFBench v1 Appendix full results; 26 isolated tasks across six categories; temperature0.1, task-specific64–2048token budgets; native fractions/errors. Paired structure+field and Field-only contexts separate. Closed-loop success uses expert sign/magnitude constraints, not physical resimulation; Human Score mean of six dimensions/two experts on120samples. Source-disputed summary/appendix cells and unbound checkpoint variants withheld.
Metric: Syntax correctness. Source: arxiv.org. Saturation forecast: Not forecast. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 VL 7B Instruct | 1 | #650 |
| 2 | Claude 3.5 Sonnet | 1 | #362 |
| 3 | Qwen 2.5 VL 72B Instruct | 1 | #383 |
| 4 | Qwen 2.5 VL 32B Instruct | 1 | #463 |
| 5 | Claude Opus 4.5 (Thinking) | 0.99 | #87 (Claude Opus 4.5) |
| 6 | GPT-4o Mini | 0.99 | #598 |
| 7 | Gemini 3 Flash | 0.99 | #103 |
| 8 | O3 | 0.99 | #132 |
| 9 | Claude Opus 4.5 (Non-reasoning) | 0.99 | #87 (Claude Opus 4.5) |
| 10 | Claude Sonnet 4.5 (Thinking) | 0.98 | #148 (Claude Sonnet 4.5) |
| 11 | GPT-5 | 0.98 | #105 |
| 12 | Claude Sonnet 4.5 (Non-reasoning) | 0.98 | #148 (Claude Sonnet 4.5) |
| 13 | GPT-4.1 | 0.97 | #258 |
| 14 | GPT-5 Mini | 0.97 | #193 |
| 15 | O4 Mini | 0.95 | #185 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=photonic-pcfbench-refactor-syntax · How It Works · Data refreshed daily, snapshot 2026-11-05.