GPTCrunch

Compare Models

Select up to 4 models to compare benchmarks, pricing, and capabilities side by side.

DeepSeek logoDeepSeek-V3.2

DeepSeek

DeepSeek logoDeepSeek-Math V2

DeepSeek

DeepSeek logoDeepSeek V4

DeepSeek

Add Model
MMLU
DeepSeek-V3.2
91.0
DeepSeek-Math V2
84.0
DeepSeek V4
91.5
HumanEval
DeepSeek-V3.2
92.0
DeepSeek-Math V2
0.0
DeepSeek V4
98.0
GSM8K
DeepSeek-V3.2
96.0
DeepSeek-Math V2
98.0
DeepSeek V4
96.0
GPQA
DeepSeek-V3.2
76.0
DeepSeek-Math V2
55.0
DeepSeek V4
80.0
SWE-bench
DeepSeek-V3.2
70.0
DeepSeek-Math V2
0.0
DeepSeek V4
75.0
MATH
DeepSeek-V3.2
92.0
DeepSeek-Math V2
97.5
DeepSeek V4
94.0
AIME 2025
DeepSeek-V3.2
88.0
DeepSeek-Math V2
85.0
DeepSeek V4
0.0
LiveCodeBench
DeepSeek-V3.2
0.0
DeepSeek-Math V2
0.0
DeepSeek V4
86.0
ModelInputOutputBlended*
DeepSeek-V3.2
$0.28$0.42$0.35
DeepSeek-Math V2
$0.27$1.10$0.69
DeepSeek V4
$0.10$0.40$0.25

*Blended = average of input and output price

Spec
DeepSeek-V3.2
DeepSeek-Math V2
DeepSeek V4
Context Window128K128K1.0M
Max Output16KN/A16K
TTFT300msN/A400ms
Speed90 tok/sN/A80 tok/s
Parameters685B (37B active)671B total / 37B active1T (estimated)
ArchitectureTransformer (MoE)MoE TransformerTransformer (MoE) + Innovations
Open SourceYesYesYes
Tierfrontierfrontierfrontier

Quick Verdict

Best Performance

DeepSeek V4

Best Value

DeepSeek V4

Fastest

DeepSeek-V3.2