PCB-Bench - Task 3 BERTScore: leaderboard
PCB-Bench task-3 free-form design-explanation generation scored by BERTScore similarity to the reference; the metric barely separates models.
Metric: BERTScore (%). Source: digailab.github.io. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B (Thinking) | 82.93 | #238 (Qwen 3 VL 235B A22B) |
| 2 | InternVL3-78B | 82.67 | #352 |
| 3 | Qwen 3 VL 8B Instruct | 82.42 | #420 |
| 4 | Llama 4 Maverick | 82.33 | #464 |
| 5 | Gemini 2.5 Pro | 82.09 | #152 |
| 6 | GPT-5 Nano | 81.99 | #425 |
| 7 | Gemini 2.5 Flash Lite | 81.94 | #422 |
| 8 | GPT-5 | 81.56 | #98 |
Interactive version: theaggregate.ai/benchmark?slug=pcb-bench-task-3-bertscore · How It Works · Data refreshed daily, snapshot 2026-10-20.