Inference
EN

DeepSeek V4 Pro 0423 vs DeepSeek V4 Flash 0731

Change models
DeepSeek V4 Pro 0423DeepSeek
DeepSeek V4 Flash 0731DeepSeek
IndexAll categories, out of 100.
57.6 ±4.8
61.9 ±3.8 best
GapIndex points below the top.
−4.3 · no clear gap
Top
CoverageIndex weight that has results.
85%
85%
SpeedOutput tokens per second.
89/s best
61/s
EloLMArena rating and rank.
1451
—
ReleasedMonth of release.
2026-04
2026-07
Input / 1MUS dollars per 1M input tokens.
$0.892
$0.043 best
Output / 1MUS dollars per 1M output tokens.
$1.78
$0.30 best
Batch output / 1MBatch price per 1M output tokens.
—
$0.144
1K chats2K tokens in, 500 out, each.
$2.68
$0.237 best
1K documents20K tokens in, 1K out, each.
$19.63
$1.17 best
1K agent tasks200K tokens in, 20K out, each.
$214.15
$14.68 best
ContextMaximum tokens in one request.
1.05M
1.05M
AgenticMulti-step tasks with tools.
55.9
59.9 best
CodingCode writing and repair.
59.2
62.8 best
ReasoningLogic problems and puzzles.
54.6
65.5 best
MultimodalTasks with images and text.
—
—
KnowledgeFacts and expert knowledge.
59.9 best
58.6
MultilingualTasks in many languages.
—
—
InstructionTasks with strict prompt rules.
52.7
59.2 best
MathMath problems.
57.6
61.5 best

Benchmarks

34 shared
Only DeepSeek V4 Pro 0423: 7Only DeepSeek V4 Flash 0731: 51
DeepSeek V4 Pro 0423
DeepSeek V4 Flash 0731
GPQA diamond · Epoch AIKnowledge
89.6%44 / 217
91.0% best33 / 217
OTIS Mock AIME 2024-2025Math
96.7% best36 / 200
94.4%47 / 200
LegalBenchLegal
80.3% best87 / 149
77.7%107 / 149
TaxEval v2Finance
72.1% best68 / 143
70.7%88 / 143
LiveCodeBenchCoding
87.5% best13 / 142
87.3%16 / 142
GPQA Diamond · Vals AIKnowledge
89.4%30 / 135
89.9% best27 / 135
MMLU ProKnowledge
87.2% best32 / 135
86.2%45 / 135
CorpFin v2Finance
61.4%52 / 131
61.8% best51 / 131
Vibe Code Bench v1.1Coding
49.9%66 / 123
74.7% best43 / 123
MedScribeHealthcare
75.1%81 / 109
80.4% best57 / 109
MedCodeHealthcare
40.5%64 / 106
41.4% best53 / 106
SimpleQA VerifiedKnowledge
47.0% best37 / 91
33.6%63 / 91
Chess PuzzlesReasoning
19.0%68 / 90
32.0% best49 / 90
FrontierMath-Tiers-1-3-v2-PrivateMath
45.3%58 / 90
57.5% best46 / 90
Legal Research BenchLegal
23.1%66 / 90
30.3% best54 / 90
SWE-benchCoding
77.4%35 / 87
88.8% best11 / 87
Tax Agent BenchFinance
25.2%44 / 82
28.1% best38 / 82
LiveBench Data AnalysisReasoning
74.5%41 / 80
79.3% best15 / 80
LiveBench ReasoningReasoning
82.7%51 / 80
86.6% best32 / 80
Mystery Game PuzzlesReasoning
17.0%51 / 80
34.0% best25 / 80