Qwen: Qwen3.8 Max Prime

Alibaba

flagship

Capabilities

VisionToolsStructured OutputExtended ThinkingLong ContextCode

Strengths

Exceptional vision and video understanding (0.98 benchmark fit)
1M token context enables extensive document and video processing
Strong math and structured output capabilities
Multimodal reasoning across text, image, and video
Extended thinking and agentic task support

Weaknesses

Flagship pricing ($12/M output) limits cost-sensitive use cases
Very slow inference (0.02 speed score, ~37 tok/s)
Closed-source with limited transparency into training or methodology
No open-weight variant for on-premise or privacy-critical deployments

Pricing

Input / 1M tokens

Output / 1M tokens

$12

Context window

1,000k

Transparency

Open weights

0.0 / 10

Open training data

0.0 / 10

Open methodology

3.0 / 10

Licence openness

1.0 / 10

Provider disclosure

4.0 / 10

FMTI company score

N/A

Composite:6.5 / 10

Alibaba/Qwen publishes minimal training methodology and does not release model weights. Transparency is limited to capability announcements and basic documentation. No open licensing or auditable training data disclosure.

Sustainability

Inference energy

N/A

Training footprint

N/A

Provider infrastructure

5.0 / 10

Composite:5.0 / 10

Alibaba operates large cloud infrastructure but has not publicly committed to renewable energy targets or disclosed carbon metrics for Qwen inference. No specific sustainability initiatives communicated.

Task Fitness

question answering

80%

code

87%

mathematics

95%

business communication

92%

vision

98%

analysis

87%

extraction

95%

long-form generation

93%

research

90%

reasoning

82%

summarisation

92%

translation

90%

conversation

92%