Qwen2.5-32B-Instruct vs Qwen2.5-72B-Instruct

Best cell per row highlighted. Null means undisclosed — never counted as zero.

Alibaba Cloud
Qwen2.5-32B-InstructGA
7.2
AI Panel
Alibaba Cloud
Qwen2.5-72B-InstructGA
7.5
AI Panel

Identity & lifecycle

ProviderAlibaba CloudAlibaba Cloud
Family / tierQwen2.5 · LargeQwen2.5 · Large
StatusGAGA
Released2024-09-182024-09-18
Knowledge cutoff2024-062024-06

Architecture & context

Context window131K131K
Max output tokens8K8K
Input modalitiestexttext
Reasoning modenonenone
Open weightsYesYes

Pricing (per Mtok)

Input$0.1$0.12
Output$0.25$0.3
Cached input
Batch input
Free tierYesYes

Speed

Output speed (tok/s)
Time to first token (s)
Latency tierfastfast

Trust & deployment

Trains on inputsNoNo
LicenseApache-2.0Qwen
CloudsGCPGCP
Compliance

AI Panel scoring

Unified score7.27.5
Decision Maker77.5
Domain Strategist77
Finance Lead99
Domain Practitioner88
Power User6.57
Skeptic7.57.5
Value score8.58.5

Benchmarks

MMLU83.386.1
MMLU-Pro6971.1
MATH-50083.183.1
HumanEval88.486.6

Scores link to their sources. Missing cells mean the vendor hasn't disclosed a result — honesty over padding.