Qwen 4 Max: benchmark results

Released Qwen4 flagship with native vision,1M context and controllable low/high/xhigh reasoning (high default).2.4T backbone plus328B learned N-gram embeddings gives2.728T under the all-parameter convention;71B backbone parameters activate per token. API-only release; smaller open models are future announcements. Provider: Alibaba. Released 2026-11-16. Access: API.

Aggregate Rating 1874 ± 11, rank #30 of 10556 rated models, from 172 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
AA APEX-Agents59.14Pass@1 (%)100
AA GDP.pdf - Insurance - Criterion Pass Rate91.75Criterion Pass Rate (%)100
EnterpriseOps-Gym-AA - Calendar53.85Task Success (%)100
LLM Stats (APEX-Agents)72.1Score (%)100
LLM Stats (Job Bench)68.5Score (%)100
Spring Prompt Deckbench - Deck Rating1585.4rating100
AA GDP.pdf - Legal - Criterion Pass Rate88.15Criterion Pass Rate (%)99.2
EnterpriseOps-Gym-AA - Teams52Task Success (%)99
AA GDPval1808.69ELO98.1
AA-Briefcase - Analytical Quality Elo1994.05Elo97.8
AA-Briefcase - Presentation Elo1651.51Elo97.8
AutomationBench-AA - Finance - Raw Objective Completion94.23Raw Objective Completion (%)97.7

Interactive version: theaggregate.ai/model?slug=qwen-4-max · How It Works · Data refreshed daily, snapshot 2026-11-17.