DeepSeek V4

Release Date: 4/23/2026

Vals Index

Accuracy (Vals Index)

55.62% ± 1.69

Latency (Vals Index)

1319.05s

Cost/Test (Vals Index)

$0.83

Context Window

1M

Max Output Tokens

384k

Input Modality

Hyperparameter settings
Default Provider : DeepSeek

Temperature

1

Top P

Default

Top K

Default

Max Output Tokens

384,000

Reasoning Effort

max

Show rankings only among open weight models

Benchmarks

Accuracy

Rankings

0.0%

± 1.69
10/ 30

0.0%

± 4.04
10/ 23

0.0%

± 0.96
38/ 116

0.0%

± 0.65
12/ 28

0.0%

± 2.93
9/ 13

0.0%

± 2.12
31/ 68

0.0%

± 2.00
42/ 65

0.0%

± 3.00
29/ 42

0.0%

± 1.26
3/ 12

0.0%

± 0.88
50/ 122

0.0%

± 4.77
16/ 66

0.0%

± 1.43
6/ 14

0.0%

± 1.65
15/ 116

0.0%

± 7.14
9/ 55

0.0%

± 0.95
6/ 122

0.0%

± 0.47
60/ 119

0.0%

± 0.34
19/ 115

0.0%

± 0.00
17/ 24

0.0%

± 4.60
8/ 12

0.0%

± 1.87
15/ 64

0.0%

± 1.50
25/ 35
Contact us
Or send us an email at contact@vals.ai

License type:

Proprietary (contact us to get access)
Industry Partner
Academic

Read our methodology.