Qwen3-30B-A3B-Base: benchmark results
Alibaba's Qwen3-30B-A3B-Base, the pretrained (non-instruct) 30B mixture-of-experts checkpoint (3B active) in the Qwen3 series, trained on 36T tokens across 119 languages. Provider: Alibaba. Released 2025-04-28. Access: Open.
Aggregate Rating 1555 ± 40, rank #597 of 1635 rated models, from 31 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Swallow - Pre-trained English - MATH | 63.4 | Exact Match (%) | 100 |
| Swallow - Pre-trained English - Average | 69.9 | Average Score (%) | 95.9 |
| Swallow - Pre-trained English - HumanEval | 69.4 | Pass@1 (%) | 95.9 |
| Swallow - Pre-trained English - MMLU | 81.2 | Accuracy (%) | 95.9 |
| Swallow - Pre-trained Japanese - JHumanEval | 64.1 | Pass@1 (%) | 95.9 |
| Swallow - Pre-trained Japanese - JMMLU | 74.3 | Accuracy (%) | 95.9 |
| Swallow - Pre-trained English - BBH | 80.4 | Exact Match (%) | 91.8 |
| Swallow - Pre-trained English - GSM8K | 82.8 | Exact Match (%) | 91.8 |
| Swallow - Pre-trained Japanese - MGSM | 74 | Exact Match (%) | 88.8 |
| Swallow - Pre-trained English - OpenBookQA | 41.4 | Accuracy (%) | 87.8 |
| Swallow - Pre-trained Japanese - Average | 58 | Average Score (%) | 87.8 |
| Swallow - Pre-trained Japanese - XL-Sum | 22.3 | ROUGE-2 (%) | 87.8 |
Interactive version: theaggregate.ai/model?slug=qwen3-30b-a3b-base · How It Works · Data refreshed daily, snapshot 2026-11-05.