Inference
EN
DeepSeek

DeepSeek V4 Flash 0731

DeepSeek V4 Flash 0731 is a reasoning model from DeepSeek in the DeepSeek V4 Flash family. 51 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-07
RankedEnough results for an index.
IndexAll categories, out of 100.61.9±3.8Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.61/sMedian 67/s
First tokenSeconds to the first output token.0.5 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.043batch $0.048Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.30batch $0.144 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.—Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
deepseek/deepseek-v4-flash-0731
Max outputMaximum output tokens per request.
944K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.035
WeightsLink to the published weights.
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
15 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
Deep InfraHost$0.06$0.181.05M
VultrDeprecatedCloud$0.10$0.251.05M
BasetenHost$0.13$0.261.05M
CoreWeaveHost$0.13$0.27262K
Arcee AIHost$0.14$0.281.05M
DigitalOceanCloud$0.14$0.281.05M
Together AIDeprecatedHost$0.14$0.281.05M
OpenRouterRouter$0.043$0.301.05M
SiliconFlowHost$0.22$0.661M
ScalewayCloud$0.468$0.936256K
Alibaba CloudHost$0.44$1.321M
ChutesHost$0.44$1.321.05M
Cloudflare Workers AIDeprecatedCloud$0.44$1.321.05M
Novita AIHost$0.44$1.321.05M
Volcengine ArkCloud$0.445$1.341M

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Run locally

304.18B parameters
16-bit
639.8 GB
8-bit
340.4 GB
4-bit
194.6 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy needs more than 128 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.17 counted59.967 / 218
CodingCode writing and repair.13 counted62.854 / 209
ReasoningLogic problems and puzzles.7 counted65.543 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted58.673 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted59.2—
MathMath problems.7 counted61.545 / 168

Results

51 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
LiveCodeBenchCoding87.3% ±1.016 / 142
SWE-benchCoding88.8% ±1.411 / 87
Harvard-MIT Mathematics Tournament February 2026Math94.8%4 / 27
Artificial Analysis GPQA DiamondKnowledge90.8%30 / 199
GPQA diamond · Epoch AIKnowledge91.0% ±1.833 / 217
Massive Multitask Language Understanding ProfessionalKnowledge86.2%12 / 75
LiveBench Data AnalysisReasoning79.3%15 / 80
GPQA Diamond · Vals AIKnowledge89.9% ±1.727 / 135
Artificial Analysis Long Context ReasoningReasoning79.7%45 / 217
Software Engineering Benchmark VerifiedCoding79.0%24 / 114
VulcanBench v3Coding88.4%3 / 14
Critical Physics TasksReasoning16.6%46 / 214
OTIS Mock AIME 2024-2025Math94.4% ±2.147 / 200
Artificial Analysis Humanity's Last ExamKnowledge38.6%53 / 222
Graduate-Level Google-Proof Q&AKnowledge88.1%37 / 141
GDPval-AA normalizedAgentic46.7%35 / 133
ARC-AGI-2 (semi-private)Reasoning61.4%36 / 131
Artificial Analysis Omniscience AccuracyKnowledge40.4%49 / 164
NL2RepoCoding54.2%11 / 36
ARC-AGI-1 (semi-private)Reasoning89.0%39 / 126
Mystery Game PuzzlesReasoning34.0% ±4.825 / 80
Terminal-Bench 2.1 · BenchLMAgentic82.7%27 / 86
MMLU ProKnowledge86.2% ±0.345 / 135
Vibe Code Bench v1.1Coding74.7% ±3.443 / 123
LiveBench ResearchAgentic79.2%19 / 52
IMOAnswerBenchMath88.4%4 / 10
LiveBench ReasoningReasoning86.6%32 / 80
Code MigrationCoding38.6% ±4.432 / 77
Artificial Analysis SciCodeCoding50.3%51 / 114
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Terminal-Bench 2.1 · Vals AIAgentic67.0% ±1.436 / 75
FrontierMath-Tiers-1-3-v2-PrivateMath57.5% ±2.946 / 90
LiveBench LanguageKnowledge79.2%34 / 66
LiveBench Long AgenticAgentic36.6%27 / 51
Humanity's Last ExamKnowledge34.8%41 / 77
GPQA Diamond · BenchLMKnowledge88.1%45 / 84
ProofBench v1.1Math56.0% ±5.029 / 54
Chess PuzzlesReasoning32.0% ±4.749 / 90
Terminal-Bench 2.0 · BenchLMAgentic56.9%23 / 41
Terminal-Bench 4.0Agentic18.7% ±3.536 / 61
ToolathlonAgentic47.8%13 / 22
LiveBench Instruction FollowingInstruction65.5%40 / 66
SkillsBenchCoding50.7% ±4.422 / 35
MCP AtlasAgentic69.0%33 / 52
BrowseCompAgentic73.2%40 / 63
FrontierMath-Tier-4-v2-PrivateMath24.4% ±6.846 / 72
LiveBench Coding 2.0Coding77.0%35 / 52
OpenHarmony Bench v1.0Coding53.8%9 / 13
SimpleQA VerifiedKnowledge33.6% ±1.563 / 91
LiveBench MathematicsMath86.8%46 / 66
LiveBench DesignAgentic71.7%37 / 52
CyberGymAgentic76.7%21 / 29
LiveBench CodingCoding75.0%49 / 66
LiveBench Agentic CodingAgentic46.8%49 / 66
SWE-bench ProCoding52.6%63 / 83
Artificial Analysis Tau3-BankingAgentic39.4%17 / 22
Toolathlon-VerifiedAgentic70.3%18 / 23
DeepSWEAgentic54.4%48 / 59
Humanity's Last Exam with toolsAgentic45.1%31 / 36
IOICoding32.7% ±2.440 / 45
Agents' Last ExamAgentic25.2%20 / 21
AutomationBenchAgentic25.1%32 / 32
ARC-AGI-2 (public eval)Reasoning63.9%33 / 122
ARC-AGI-1 (public eval)Reasoning94.8%35 / 116
CyberBench v1.1Other72.1% ±5.418 / 48
CorpFin v2Finance61.8% ±1.051 / 131
Tax Agent BenchFinance28.1% ±3.238 / 82
MedCodeHealthcare41.4% ±2.153 / 106
MedScribeHealthcare80.4% ±2.057 / 109
ProgramBenchCoding0.0%34 / 65
EMBFinance57.0% ±3.442 / 75
Terminal-Bench-ScienceScience4.3% ±2.426 / 44
Finance Agent (v2)Finance49.5% ±0.547 / 79
Legal Research BenchLegal30.3% ±3.254 / 90
TaxEval v2Finance70.7% ±0.988 / 143
LegalBenchLegal77.7% ±0.5107 / 149
BioMysteryBenchScience64.4% ±1.324 / 30
LiveCodeBench Pass@1 with Chain-of-ThoughtCoding91.6%—
Apex ShortlistMath85.7%—
Chinese-SimpleQAKnowledge78.9%—
MRCR 1MReasoning78.7%—
DeepSeek DSBench FullStackCoding68.7%—
CorpusQA 1MReasoning60.5%—
DeepSeek DSBench HardCoding59.6%—
ApexMath33.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYVals AINo licence statedARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedmodels.devMIT

History

5 releases
Index62.8 → 61.9
23 Oct 202612 Nov 2026
Output / 1M$0.45 → $0.30
23 Oct 202612 Nov 2026

Questions

How much does DeepSeek V4 Flash 0731 cost?
$0.043 per 1M input tokens, $0.30 per 1M output tokens.
How good is DeepSeek V4 Flash 0731?
Its index is 61.9 out of 100. That is place 65 of 283 ranked models.
What is DeepSeek V4 Flash 0731 best at?
Reasoning: 65.5 out of 100, place 43 of 227.
How fast is DeepSeek V4 Flash 0731?
61 output tokens per second, 0.5 s to the first token.
What is the context window of DeepSeek V4 Flash 0731?
1.05M tokens.
Is DeepSeek V4 Flash 0731 open source?
Yes. Its weights are published.
When was DeepSeek V4 Flash 0731 released?
July 2026.

Badge

DeepSeek V4 Flash 0731 on Inference360: place 65 of 283, index 61.9
HTML
<a href="https://inference360.ai/models/deepseek-v4-flash-0731"><img src="https://inference360.ai/badges/deepseek-v4-flash-0731.svg" alt="DeepSeek V4 Flash 0731 on Inference360: place 65 of 283, index 61.9" height="20"></a>
Markdown
[![DeepSeek V4 Flash 0731 on Inference360: place 65 of 283, index 61.9](https://inference360.ai/badges/deepseek-v4-flash-0731.svg)](https://inference360.ai/models/deepseek-v4-flash-0731)

Compare with

GPT-5.3 Codex62.0Claude Opus 4.762.0Gemini 3.1 Pro Preview61.5Mistral Large 460.7dots3-note Preview63.4

Lower price

dots3-note Preview63.4 · FreeGLM-5.3-Flash63.1 · $0.078/1MLing 3.1 Flash67.2 · $0.135/1M

More from DeepSeek

All models
DeepSeek V4 Flash LatestUnrankedDeepSeek V4 Pro 081365.5DeepSeek V4 Flash Vision Exp60.4DeepSeek Flash LatestUnrankedDeepSeek Pro LatestUnranked