DeepSeek V4

Release Date: 4/23/2026

Vals Index

Accuracy (Vals Index)

63.87% ± 1.94

Latency (Vals Index)

435.62s

Cost/Test (Vals Index)

$0.32

Context Window

1M

Max Output Tokens

384k

Input Modality

Hyperparameter settings
Default Provider : DeepSeek

Temperature

1

Top P

0.95

Top K

Default

Max Output Tokens

384,000

Reasoning Effort

max

Show rankings only among open weight models

Benchmarks

Accuracy

Rankings

0.0%

± 1.94
9/ 45

0.0%

± 1.50
26/ 50

0.0%

± 0.96
27/ 100

0.0%

± 2.80
4/ 49

0.0%

± 2.12
26/ 54

0.0%

± 2.00
35/ 54

0.0%

± 3.02
19/ 28

0.0%

± 0.88
42/ 107

0.0%

± 4.77
9/ 40

0.0%

± 1.65
9/ 102

0.0%

± 7.14
6/ 52

0.0%

± 0.95
3/ 108

0.0%

± 0.47
51/ 119

0.0%

± 0.34
15/ 101

0.0%

± 1.87
8/ 44

0.0%

± 5.29
11/ 56
Contact us
Or send us an email at contact@vals.ai
Proprietary Benchmarks (contact us to get access)
Academic Benchmarks

Read about our methodology.