QwQ-32B vs Grok 4.20

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

Alibaba Cloud
QwQ-32BGA
6.8
AI Panel
xAI
Grok 4.20GA
6.9
AI Panel

Identity & lifecycle

ProviderAlibaba CloudxAI
Family / tierQwQ · ReasoningGrok 4 · Reasoning
StatusGAGA
Released2025-03-042026-03-09
Knowledge cutoff2024-092024-11

Architecture & context

Context window131K1M
Max output tokens33K
Input modalitiestexttext, image
Reasoning modealwaysoptional
Open weightsYesNo

Pricing (per Mtok)

Input$0.12$1.25
Output$0.18$2.5
Cached input$0.2
Batch input
Free tierYesYes

Speed

Output speed (tok/s)171.4
Time to first token (s)13.24
Latency tierslowmedium

Trust & deployment

Trains on inputsNoYes
LicenseApache-2.0Proprietary
CloudsGCP
Compliance

AI Panel scoring

Unified score6.86.9
Decision Maker77
Domain Strategist76.5
Finance Lead76.5
Domain Practitioner7.57.5
Power User66.5
Skeptic6.56
Value score87

Benchmarks

GPQA Diamond65.278.5
MATH-50090.687.3
LiveCodeBench63.4
IFEval83.981
TAU-bench93
LMArena Elo1491
Artificial Analysis Index49

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