Compare Models
DeepSeek V4 Flash 0731

Release Date: Jul 31, 2026

Developer DeepSeek 🇨🇳
Context Window 1M
Max Output Tokens 384k
Token Costs (in/out) $0.44/1.32
Weights Open
Input Modalities

Accuracy

53.57 % ± 1.21

Cost / Test (Vals Index)

$ 0.866

Latency

34 min 55 s

Vals Index
BenchmarksAccuracyRankings
Vals Index

0.0%

±1.21
23/57
Code Migration

0.0%

±4.38
18/59
CyberBench

0.0%

±5.59
6/25
EMB

0.0%

±3.39
28/57
Finance Agent (v2)

0.0%

±0.47
29/60
Legal Research Bench

0.0%

±3.19
30/60
MedCode

0.0%

±2.15
42/92
MedScribe

0.0%

±1.97
45/94
ProofBench v1.1

0.0%

±4.96
12/31
TaxEval v2

0.0%

±0.90
90/145
Vibe Code Bench v1.1

0.0%

±3.44
19/95
BioMysteryBench

0.0%

±1.28
8/11
Harvey's Legal Agent Benchmark

0.0%

±2.18
14/61
GPQA Diamond

0.0%

±1.69
27/138
IOI

0.0%

±2.35
23/27
LiveCodeBench

0.0%

±0.97
17/143
LegalBench

0.0%

±0.50
106/144
MMLU Pro

0.0%

±0.34
47/138
ProgramBench

0.0%

±0.00
22/45
SkillsBench

0.0%

±4.43
22/35
SWE-bench

0.0%

±1.41
11/88
Terminal-Bench 4.0

0.0%

±0.88
17/27
Terminal-Bench Science

0.0%

±0.00
23/25
Proprietary BenchmarksAcademic BenchmarksIndustry Partners
Vals
Default Provider : DeepSeek
Temperature: 1
Top P: Default
Top K: Default
Max Output Tokens: 384,000
Reasoning Effort: high