Qwen3 Max

Name: Qwen3 Max
Author: Alibaba

Alibaba's >1T-parameter dense Qwen3 flagship, available only as a closed API on Qwen Chat and Alibaba Cloud.

Open

Publisher: Alibaba
Family: Qwen3
Params: >1T (closed)
Context: 262K
Input
Output
Providers: alibaba-cloud
License: proprietary
Released: Sep 2025
Canonical: qwen.ai

Cite

Notes

Only stored in your browser.

Attribution

Benchmark scores: AA prime-hub AfterQuery Vals AI EQ-Bench

Attribution policy →

Context

262K

Weights

Closed

API only

Input

/1M tok

Output

/1M tok

Blended

#1/ 500/1M · 3:1

Speed

#106/ 499tok/s

Cache read

$0.16

/1M tok

Cache write

$0.97

/1M tok

Price & speed via Artificial Analysis · ranked across tracked models · caching via OpenRouter

Reported on 18 evals across 11 domains - best 84.1% on MMLU-Pro · #39 of 340

Reported eval scores

CorpFin v268

TaxEval v273.5

MMLU-Pro84.1

Humanity's Last Exam (HLE)11.9

LiveCodeBench76.7

AIME 2025: Problems from the American Invitational Mathematics Examination80.7

IDE-Bench65

Vibe Code Bench v1.13.5

τ²-bench (Tau²-bench)74.3

GPQA Diamond76.4

SciCode38.3

Terminal-Bench (Hard)20.5

IFBench44.1

EQ-Bench Longform Creative Writing66.2

Physgym Arena Drhard Public45.2

Physgym Arena Medley Public80.4

MedScribe72.7

MedCode31.4