Qwen 3 30B A3B 2507 Instruct — benchmark results

Alibaba's July 2025 refresh of its open 30B A3B MoE (3.3B active), non-thinking instruct variant with 256K native context. Provider: Alibaba. Released 2025-07-01. Access: Open.

Unified ELO 1544 ± 5, rank #639 of 1776 rated models, from 729 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
LIBRA - LibrusecHistory96.12Dataset Total Score (%)100
LIBRA - LibrusecMHQA *51Dataset Total Score (%)100
LIBRA - MatreshkaNames *81.2Dataset Total Score (%)100
LIBRA - ru2WikiMultihopQA *66.63Dataset Total Score (%)100
LIBRA - ruBABILongQA180.5Dataset Total Score (%)100
LIBRA - ruBABILongQA264.72Dataset Total Score (%)100
LIBRA - ruBABILongQA3 *38.38Dataset Total Score (%)100
LIBRA - ruBABILongQA470.98Dataset Total Score (%)100
LIBRA - ruBABILongQA582.95Dataset Total Score (%)100
LIBRA - ruSciAbstractRetrieval81.5Dataset Total Score (%)100
LIBRA - ruSciFi60.6Dataset Total Score (%)100
LIBRA - ruSciPassageCount *25.77Dataset Total Score (%)100

Interactive version: theaggregate.ai/model?slug=qwen-3-30b-a3b-2507-instruct · How the rankings work · Data refreshed daily, snapshot 2026-07-22.