AlibabaProprietary

Qwen3-Max-Thinking

Qwen3-Max-Thinking is an inference model developed by Alibaba. It has a large-scale configuration of approximately 10 trillion parameters and a very long context window of 1 million tokens.

Parameters

10000.0B

Context Window

1000K

License

Proprietary

Release Date

2026-01-26

API Pricing

API pricing for this model is not yet available

Strengths

・Overwhelming parameter scale
・1 million token long-context comprehension
・Pursuing advanced reasoning capabilities

Weaknesses

・Non-public closed model
・Limited license
・Detailed performance metrics not disclosed

Use Cases

・Executing complex logical reasoning
・Analyzing ultra-large-scale data
・Automating advanced problem-solving

Deep Analysis

Release Date

January 23, 2026

Parameters

Proprietary (undisclosed)

Context Window

262,144 tokens

Architecture

Decoder-only with extended thinking

Input Price

$0.78/1M tokens

Output Price

$3.90/1M tokens

GPQA Diamond

87.4

SWE-bench Verified

75.3

HLE (w/ tools)

49.8

API Model Name

qwen3-max-2026-01-23

Strengths

・Competitive with GPT-5.2-Thinking and Claude-Opus-4.5 on 19 established benchmarks
・Adaptive tool-use: autonomously invokes Search, Memory, and Code Interpreter
・Excellent value: $0.78/$3.90 pricing is much cheaper than GPT-5.2 and Claude Opus 4.5
・100% reliability rate across evaluated benchmarks — never fails to produce output
・Strong on C-Eval (93.7), HLE with tools (49.8), and coding tasks (97th percentile)

Weaknesses

・General knowledge is a notable weakness (23rd percentile on broad factual recall)
・Trails GPT-5.2-Thinking on MMLU-Pro (85.7 vs 87.4) and GPQA (87.4 vs 92.4)
・Now partially superseded by Qwen3.7-Max as the flagship reasoning model
・Test-time scaling adds latency and token cost for heavy mode
・SWE-bench 75.3% trails Claude Opus 4.5 (80.9%) and GPT-5.2 (80.0%)

Competitor Comparison

Model	Arena	SWE	GPQA	Price
GPT-5.2-Thinking	~1500	80.0	92.4	Proprietary
Claude-Opus-4.5	~1490	80.9	87.0	Proprietary
Gemini 3 Pro	~1480	76.2	91.9	Proprietary
Qwen3-Max-Thinking	~1450	75.3	87.4	$0.78/$3.90
DeepSeek V3.2	~1430	73.1	82.4	Proprietary

Sources

Analysis generated: 2026-05-24