Qwen 2.5 32B Instruct: benchmark results

Alibaba Qwen 2.5 32B instruction-tuned checkpoint. Provider: Alibaba. Released 2024-09-19. Access: Open.

Aggregate Rating 1583 ± 5, rank #498 of 1656 rated models, from 902 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AbstentionBench - underspecified context - ALCUNA - F183.25Abstention F1 (%)100
AbstentionBench - underspecified context - ALCUNA - Recall84.06Abstention Recall (%)100
AbstentionBench - underspecified context - BBQ - F195.78Abstention F1 (%)100
AbstentionBench - underspecified context - BBQ - Recall95.05Abstention Recall (%)100
AbstentionBench - underspecified context - Musique - Recall87.08Abstention Recall (%)100
AbstentionBench - underspecified context - SQuAD 2.0 - F184.13Abstention F1 (%)100
AbstentionBench - underspecified context - SQuAD 2.0 - Recall74.99Abstention Recall (%)100
FinTradeBench (RAG) - Fundamentals47Normalized correctness: a calibrated Claude Sonnet 4.5 judge100
FinTradeBench - Fundamentals42.3Normalized correctness: a calibrated Claude Sonnet 4.5 judge100
KOR-Bench - Operation93.2Accuracy (%; zero-shot, 250 questions)100
MedStruct-S (Two-Shot) - Semi-Structured Extraction49.24Key-value pair F1 (%; exact key and exact value, micro-avera100
SeaEval - Fundamental NLP Tasks - MNLI (Zero-Shot)87.15Accuracy (%)100

Interactive version: theaggregate.ai/model?slug=qwen-2-5-32b-instruct · How It Works · Data refreshed daily, snapshot 2026-10-20.