Models/Qwen3 Max Thinking (Preview)

MDL

Qwen3 Max Thinking (Preview)

Alibaba

Verified against Artificial Analysis · Jul 13, 2026

Capability

percentile index

Price /M

$1.2 / $6

in / out

Context

—

Avg score

55.9

10 benchmarks

AA Intelligence

25.0

index

Speed

62 tok/s

TTFT 1.8s

Released

Nov 3, 2025

About

Qwen3 Max Thinking (Preview) is a open-weights model in the Qwen3 family from Alibaba. Benchmarked on 10 evals, averaging 55.9.

Benchmark scores · 10

avg 55.9 — higher is better

τ²-benchartificial-analysis · 2026-07

83.6

MMLU-Proartificial-analysis · 2026-07

82.4

AIMEartificial-analysis · 2026-07

82.3

GPQA Diamondartificial-analysis · 2026-07

77.6

AA Long-Context Reasoningartificial-analysis · 2026-07

57.7

IFBenchartificial-analysis · 2026-07

53.8

LiveCodeBenchartificial-analysis · 2026-07

53.5

SciCodeartificial-analysis · 2026-07

38.7

Terminal-Bench Hardartificial-analysis · 2026-07

17.4

Humanity's Last Examartificial-analysis · 2026-07

The research behind this model

all papers mentioning it →

IndustryBench: Probing the Industrial Knowledge Boundaries of LLMsMay 11, 2026

Mentions matched by name in title/abstract — from the arXiv + HF daily corpus.

Details

PublisherAlibaba

FamilyQwen3

Parameters—

License—

WeightsOpen

ReleasedNov 3, 2025

Performance & price

Output speed62 tok/s

Latency (TTFT)1.8s

First answer34s

Price in / out$1.2 / $6 per 1M

Blended price$2.4 / 1M

Math index82.3

Median across API providers, via Artificial Analysis. Blended = 3:1 input:output. For reasoning models, latency includes thinking time. Methodology →

History

full ledger →

Capability over time

Append-only ledger — every observed change to this model's numbers.

API providers

Artificial Analysis$6/M