2026-10-20UPDATEPlatform updated: 52 models tracked across 42 independent benchmarks including CorpFin • 2026-10-20BENCHCorpFin added to benchmark suite — long credit agreement comprehension evaluations live • 2026-10-19MODELClaude Sonnet 5.5 and Opus 5.5 achieve 75.9% and 75.8% on Excel Modeling Benchmark v1.1 • 2026-10-19MODELTax Agent Bench updated: Claude Fable 5.1 (49.2%) and Gemini 4 Argon (44.6%) lead corporate tax research • 2026-10-18MODELGemini 4 Argon claims #1 rank on Finance Agent (v2) with 65.4% • 2026-10-13MODELGoogle launches Gemini 4 Argon with 1M-token output limit and native multimodal reasoning • 2026-10-06MODELAnthropic completes Claude 5.5 generation with Claude Haiku 5.5, Opus 5.5, and Sonnet 5.5 • 2026-10-20UPDATEPlatform updated: 52 models tracked across 42 independent benchmarks including CorpFin • 2026-10-20BENCHCorpFin added to benchmark suite — long credit agreement comprehension evaluations live • 2026-10-19MODELClaude Sonnet 5.5 and Opus 5.5 achieve 75.9% and 75.8% on Excel Modeling Benchmark v1.1 • 2026-10-19MODELTax Agent Bench updated: Claude Fable 5.1 (49.2%) and Gemini 4 Argon (44.6%) lead corporate tax research • 2026-10-18MODELGemini 4 Argon claims #1 rank on Finance Agent (v2) with 65.4% • 2026-10-13MODELGoogle launches Gemini 4 Argon with 1M-token output limit and native multimodal reasoning • 2026-10-06MODELAnthropic completes Claude 5.5 generation with Claude Haiku 5.5, Opus 5.5, and Sonnet 5.5 •
Home/Models/DeepSeek V4 Flash
DeepSeek logoDeepSeek

DeepSeek V4 Flash

Open-weight#20 on FB Leaderboard1M contextReleased 2026-07-3119 benchmarks evaluated
View Benchmarks →Submit a score
63.2
FB Score
#20
Overall rank
FB Score63.2Rank #20
Best benchmark89.9%GPQA Diamond
PAVE overall—%Level 1-4
Benchmarks19 / 42Evaluated
✦Official Certification
Get your AI model certified by FinanceBenchmark
Verified performance scores · Embeddable trust badge · Leaderboard listing
Claim Your Badge

Score by domain

Verification
66.3%
Rank #22 of 33
Document QA
61.3%
Rank #36 of 45
Numerical
71.4%
Rank #21 of 32
Forensic
53.6%
Rank #30 of 45
Multi-domain
Not evaluated
No data
Agent Tasks
57.6%
Rank #6 of 44
Performance profile
80%Verification66.3%Document QA61.3%Numerical71.4%Forensic53.6%Agent Tasks57.6%
Benchmark results19 evaluated · 23 pending
BenchmarkScoreRankTierWeightSourceDate
Pave Interchange Fee Benchmark
VerificationForensic
——Tier 130FB 2026Pending
FinanceBench
Document QA
——Tier 220FB 2026Pending
FinBen
Multi-domain
——Tier 310FB 2026Pending
FinQA
Numerical
——Tier 220FB 2026Pending
ConvFinQA
Document QA
——Tier 220FB 2026Pending
Finance Agent Benchmark
Agent Tasks
49.5#30 / 44Tier 130FB 20262026-09
MultiHiertt
Numerical
——Tier 310FB 2026Pending
FinAuditing
ForensicCompliance
——Tier 310FB 2026Pending
VALS Index
VerificationNumericalAgent Tasks
53.6#12 / 22Tier 125FB 20262026-07
MortgageTax Benchmark
NumericalForensic
——Tier 220FB 2026Pending
TaxEval v2
VerificationForensic
70.7#23 / 24Tier 220FB 20262026-07
Code Migration
Agent Tasks
38.6#8 / 23Tier 310FB 20262026-07
CyberBench
Forensic
77.5#4 / 11Tier 310FB 20262026-07
Excel Modeling Benchmark
Document QAForensic
57.0#28 / 44Tier 310FB 20262026-09
Legal Research Bench
Document QAForensic
30.3#15 / 22Tier 310FB 20262026-07
MedCode
Verification
41.4#14 / 20Tier 310FB 20262026-07
MedScribe
Document QA
80.4#16 / 20Tier 310FB 20262026-07
ProofBench v1.1
NumericalVerification
56.0#9 / 13Tier 310FB 20262026-07
SAGE
Agent Tasks
——Tier 310FB 2026Pending
Public Benefits Bench
VerificationForensic
——Tier 310FB 2026Pending
Vibe Code Bench v1.1
Agent Tasks
74.7#10 / 23Tier 310FB 20262026-07
Harvey's Legal Agent
Agent TasksForensic
8.3#8 / 22Tier 310FB 20262026-07
GPQA Diamond
NumericalVerification
89.9#13 / 20Tier 215FB 20262026-07
IOI
Numerical
——Tier 310FB 2026Pending
LiveCodeBench
Agent Tasks
87.3#7 / 19Tier 310FB 20262026-07
LegalBench
Document QAForensic
77.7#21 / 21Tier 310FB 20262026-07
MMLU Pro
NumericalVerification
86.2#16 / 21Tier 215FB 20262026-07
MMMU
Document QANumerical
——Tier 310FB 2026Pending
ProgramBench
Agent Tasks
——Tier 310FB 2026Pending
SkillsBench
Agent Tasks
50.7#10 / 17Tier 310FB 20262026-07
SWE-bench
Agent Tasks
88.8#8 / 22Tier 220FB 20262026-07
Terminal-Bench 2.1
Agent Tasks
67.0#14 / 23Tier 310FB 20262026-07
RSI Index
Agent TasksNumerical
——Tier 310FB 2026Pending
Time Horizon Index
ForensicAgent Tasks
——Tier 310FB 2026Pending
ReverseEngBench
Forensic
——Tier 310FB 2026Pending
Tax Agent Bench
ForensicAgent Tasks
——Tier 310FB 2026Pending
Big Finance Bench
ForensicDocument QA
——Tier 310FB 2026Pending
GDPval-MM
Document QANumerical
——Tier 310FB 2026Pending
BankerToolBench
Agent Tasks
——Tier 310FB 2026Pending
GDPval-Rubrics
Forensic
——Tier 310FB 2026Pending
PRBench-Finance
NumericalForensic
——Tier 310FB 2026Pending
CorpFin
ForensicVerification
——Tier 220FB 2026Pending
Contents.md
Model info
OrganisationDeepSeek
Release date2026-07-31
LicenseOpen-weight
Context window1M tokens
Open weightYes
FB Score breakdown
Finance Agent×30 = 14.9
VALS Index×25 = 13.4
TaxEval v2×20 = 14.1
Code Migration×10 = 3.9
CyberBench×10 = 7.7
EMB×10 = 5.7
Legal Research×10 = 3.0
MedCode×10 = 4.1
MedScribe×10 = 8.0
ProofBench×10 = 5.6
Vibe Code Bench×10 = 7.5
Harvey Legal×10 = 0.8
GPQA Diamond×15 = 13.5
LiveCodeBench×10 = 8.7
LegalBench×10 = 7.8
MMLU Pro×15 = 12.9
SkillsBench×10 = 5.1
SWE-bench×20 = 17.8
Terminal-Bench×10 = 6.7
FB Score (weighted avg)63.2
Compare - all models
AClaude Opus 4.773.8%
OGPT-5.572.2%
OGPT-6 Sol71.1%
DDeepSeek V4 Pro70.1%
GGemini 3.1 Pro70.0%
OGPT-6 Luna70.0%
Compare vs Claude Opus 4.7 in full →
Evaluation notes
Scores reflect median of 3 runs per task at temperature 0.0 with 0-shot prompting. PAVE evaluation conducted by FinanceBenchmark. Other scores sourced from latest published arXiv papers.