Qwen3-14B-Base: benchmark results
Alibaba's Qwen3-14B-Base, the pretrained (non-instruct) dense 14B checkpoint in the Qwen3 series, trained on 36T tokens across 119 languages with a three-stage pretraining pipeline. Provider: Alibaba. Released 2025-04-29. Access: Open.
Aggregate Rating 1548 ± 16, rank #611 of 1640 rated models, from 210 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| STEMVerse - Biology - Biophysics and Biochemistry | 52.35 | Few-shot accuracy (%) | 100 |
| STEMVerse - Biology - Genetics and Bioinformatics | 55.5 | Few-shot accuracy (%) | 100 |
| STEMVerse - Chemistry - Physical Chemistry | 40.2 | Few-shot accuracy (%) | 100 |
| STEMVerse - Physics - Bloom Level: Apply | 28.56 | Few-shot accuracy (%) | 100 |
| STEMVerse - Physics - Electrodynamics | 36.61 | Few-shot accuracy (%) | 100 |
| STEMVerse - Physics - Fluid Mechanics | 32.46 | Few-shot accuracy (%) | 100 |
| STEMVerse - Physics - Thermodynamics and Statistical Physics | 28.45 | Few-shot accuracy (%) | 100 |
| Swallow - Pre-trained Japanese - JHumanEval | 70.9 | Pass@1 (%) | 100 |
| Swallow - Pre-trained English - HumanEval | 70.9 | Pass@1 (%) | 99 |
| STEMVerse - Chemistry - Bloom Level: Evaluate | 70 | Few-shot accuracy (%) | 96.7 |
| STEMVerse - Chemistry - Bloom Level: Understand | 92.31 | Few-shot accuracy (%) | 96.7 |
| Swallow - Pre-trained Japanese - XL-Sum | 26.1 | ROUGE-2 (%) | 95.9 |
Interactive version: theaggregate.ai/model?slug=qwen3-14b-base · How It Works · Data refreshed daily, snapshot 2026-10-13.