Qwen3-30B-A3B-Thinking-2507-FP8: benchmark results
Provider: Alibaba. Released 2025-07-31. Access: Open.
Aggregate Rating 1631 ± 25, rank #746 of 10916 rated models, from 45 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Neo-Classic - Word Cloze (Tone Distractors, CoT) | 73 | Accuracy (%; restore a masked two-character word in a contem | 85.7 |
| Medmarks - M-ARC | 63.33 | Score (%) | 84.1 |
| JailNewsBench (Context Overload) - Harmfulness Score | 2.1 | Mean harmfulness of the generated fake news (0 to 4): the av | 81.2 |
| JailNewsBench (Research Front) - Harmfulness Score | 2.5 | Mean harmfulness of the generated fake news (0 to 4): the av | 81.2 |
| Medmarks - Med-HALT Reasoning NOTA | 66.71 | Score (%) | 76.1 |
| JailNewsBench (Role Play) - Harmfulness Score | 2.4 | Mean harmfulness of the generated fake news (0 to 4): the av | 75 |
| Medmarks - MedHallu Medium | 63.36 | Score (%) | 73.9 |
| Medmarks - MedHallu Easy | 70.07 | Score (%) | 72.7 |
| Medmarks - SuperGPQA Medicine Easy | 58.38 | Score (%) | 70.5 |
| JailNewsBench (Negative Prompting) - Harmfulness Score | 2.3 | Mean harmfulness of the generated fake news (0 to 4): the av | 68.8 |
| Medmarks - PubMedQA | 76.6 | Score (%) | 68.2 |
| Medmarks - Med-HALT Reasoning FCT | 78.89 | Score (%) | 67 |
Interactive version: theaggregate.ai/model?slug=qwen3-30b-a3b-thinking-2507-fp8 · How It Works · Data refreshed daily, snapshot 2026-11-08.