Qwen 3 Next 80B A3B Instruct: benchmark results
Direct-answer checkpoint using hybrid attention to process long inputs efficiently. It does not emit separate thinking blocks. Provider: Alibaba. Released 2025-09-11. Access: Open.
Aggregate Rating 1627 ± 6, rank #368 of 1656 rated models, from 935 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - conceptual_design | 2.22 | Dataset z-score | 100 |
| AGC-Bench - tinystories | 1.8 | Dataset z-score | 100 |
| AdversaRiskQA | 94.7 | Accuracy (%; mean of six domain-by-difficulty accuracies on | 100 |
| CarryOnBench - Utility (Benign Clarifications) | 75.2 | Ben-Util (%): share of the benign information-need checklist | 100 |
| CarryOnBench - Utility (No Clarifications) | 49.6 | Ben-Util (%): share of the benign information-need checklist | 100 |
| ClinicalMC (Chinese) | 48.17 | Mean of 11 task scores (%; triage accuracy, examination reca | 100 |
| FoundationalASSIST - Most Common Distractor | 47.9 | Accuracy (%; which wrong option students chose most often, 2 | 100 |
| Nejumi 4 - jaster (2-shot) - JSQuAD | 82.64 | Character F1 (x100) | 100 |
| AGC-Bench - future_ideas | 1.62 | Dataset z-score | 98.8 |
| AGC-Bench - thenextchapter | 2.37 | Dataset z-score | 98.8 |
| EuroEval Spanish | 64.72 | Average Score (%) | 98.8 |
| EuroEval French | 72.45 | Average Score (%) | 98.7 |
Interactive version: theaggregate.ai/model?slug=qwen-3-next-80b-a3b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.