Model

Qwen3-Max

CLOSED

Alibaba · Qwen3-Max family · released Sep 5, 2025

Alibaba's biggest model yet, over 1T params (exact undisclosed) trained on 36T tokens; API-only preview.

ReasoningCodingVisionFunction callingTool useAgentic

1658.1

Elo · rank #92

Parameters

Undisclosed

Active params

Undisclosed

Context

256K tokens

Architecture

MoE, exact expert configuration undisclosed (>1T total params)

License

Proprietary

Languages

100+

API price (in/out)

$1.2 / $6

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

AIMEMath80.6%#75

best: GPT-5.2 · 100.0%

Arena-Hard v2Human preference86.1%#1

best: this model · 86.1%

Arena EloHuman preference1432#14

best: Claude Fable 5 · 1505

GPQA DiamondReasoning76.0%#99

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning9.0%#92

best: Claude Sonnet 5 · 57.4%

IFBenchReasoning48.0%#37

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding57.5%#102

best: DeepSeek-V4-Pro (Think Max) · 93.5%

SuperGPQAReasoning64.6%#12

best: Qwen3.7-Max · 73.6%

SWE-bench VerifiedCoding69.6%#60

best: Claude Fable 5 · 95.0%

τ²-Bench TelecomAgents74.8%#26

best: Claude Opus 4.6 · 99.3%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

$1.2

Output / M tok

Qwen3-Max family

Elo progression across releases

Qwen3-MaxSep 20251658.1

Resources

DOCSwww.alibabacloud.com/help/en/model-studio/models APIAvailable via Alibaba API

API price $1.2/$6 · each benchmark row carries its own source badge (see methodology)