Photonic PCFBench - Closed Loop Success: 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: Success rate. Source: arxiv.org. Saturation forecast: Not forecast. 31 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 0.8 | #598 |
| 2 | O3 | 0.7 | #132 |
| 3 | Claude Opus 4.5 (Thinking) | 0.69 | #87 (Claude Opus 4.5) |
| 4 | Gemini 3 Flash | 0.68 | #103 |
| 5 | Qwen 2.5 VL 7B Instruct | 0.64 | #650 |
| 6 | Claude Opus 4.5 (Non-reasoning) | 0.63 | #87 (Claude Opus 4.5) |
| 7 | GPT-5 | 0.62 | #105 |
| 8 | Qwen 3 VL 30B A3B Instruct | 0.62 | #390 |
| 9 | Gemini 2.5 Flash Lite | 0.61 | #432 |
| 10 | Qwen 2.5 VL 72B Instruct | 0.59 | #383 |
| 11 | O4 Mini | 0.56 | #185 |
| 12 | Claude Sonnet 4.5 (Thinking) | 0.55 | #148 (Claude Sonnet 4.5) |
| 13 | GPT-5 Mini | 0.54 | #193 |
| 14 | GPT-4.1 | 0.5 | #258 |
| 15 | Qwen 2 VL 72B Instruct | 0.48 | #551 |
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-closed-loop-success · How It Works · Data refreshed daily, snapshot 2026-11-05.