Qwen 3 VL 30B A3B Instruct: benchmark results

Alibaba's open 30B A3B MoE vision-language model with 256K context, OCR in 32 languages, and GUI-agent skills. Provider: Alibaba. Released 2025-07-01. Access: Open.

Unified ELO 1526 ± 1, rank #563 of 1392 rated models, from 101 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
FlagEval EmbodiedVerse - EgoPlan-Bench258.8Score100
Konkur 1404 - Experimental Sciences50.31Accuracy (%, text-only)100
LLM Stats (CharadesSTA)63.5Score (%)86.4
LLM Stats (MLVU-M)81.3Score (%)85.7
Konkur 1404 - Mathematics46.96Accuracy (%, text-only)84.2
FlagEval EmbodiedVerse - CV-Bench (test)86.78Score82.6
LLM Stats (OCRBench)90.3Score (%)82.6
Konkur 1404 - Overall43.64Accuracy (%, text-only)78.9
AA Omniscience - Software Engineering (SWE) - Swift40Accuracy (%)70.1
LLM Stats (ScreenSpot)94.7Score (%)70
FlagEval EmbodiedVerse - EmbSpatial-Bench76.29Score69.6
FlagEval EmbodiedVerse - VSI-Bench (tiny)47.49Score69.6

Interactive version: theaggregate.ai/model?slug=qwen-3-vl-30b-a3b-instruct · How It Works · Data refreshed daily, snapshot 2026-09-05.