Inference
EN
DeepSeek

DeepSeek V4 Pro 0813

DeepSeek V4 Pro 0813 is a reasoning model from DeepSeek in the DeepSeek V4 Pro family. 52 benchmarks count toward its index, in 6 categories.

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.65.5±3.7Median 51.0
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.42/sMedian 67/s
First tokenSeconds to the first output token.0.4 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.111batch $0.66Median $0.60
Output / 1MUS dollars per 1M output tokens.$4.44batch $1.98 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-pro-0813
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.
$0.097
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.
13 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
DeepSeekMaker$0.66$1.981.05M
Deep InfraHost$1.30$2.601.05M
CoreWeaveHost$1.31$3.961.05M
Arcee AIHost$1.32$3.961.05M
BasetenHost$1.32$3.961.05M
Cloudflare Workers AIDeprecatedCloud$1.32$3.961.05M
DigitalOceanCloud$1.32$3.961.05M
NebiusHost$1.32$3.96979K
Novita AIHost$1.32$3.961.05M
SiliconFlowHost$1.32$3.961M
Together AIHost$1.32$3.961.05M
Volcengine ArkCloud$1.34$4.011M
OpenRouterRouter$0.111$4.441.05M

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

Run locally

1650.5B parameters
16-bit
3467.1 GB
8-bit
1842.3 GB
4-bit
1051.6 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 counted63.750 / 218
CodingCode writing and repair.13 counted65.142 / 209
ReasoningLogic problems and puzzles.7 counted68.526 / 227
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted65.045 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted62.011 / 63
MathMath problems.7 counted64.136 / 168

Results

52 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
SWE-benchCoding96.4% ±0.82 / 87
Artificial Analysis GPQA DiamondKnowledge92.8%14 / 199
Massive Multitask Language Understanding ProfessionalKnowledge87.5%6 / 75
τ²-Bench Tool-Agent-User EvaluationAgentic96.2%12 / 149
Artificial Analysis IFBenchInstruction76.5%12 / 146
NL2RepoCoding61.5%3 / 36
LiveCodeBenchCoding87.5% ±1.013 / 142
Harvard-MIT Mathematics Tournament February 2026Math95.2%3 / 27
GPQA Diamond · Vals AIKnowledge92.4% ±2.016 / 135
Artificial Analysis Omniscience AccuracyKnowledge49.1%22 / 164
OTIS Mock AIME 2024-2025Math98.6% ±1.027 / 200
GPQA diamond · Epoch AIKnowledge91.7% ±1.530 / 217
Terminal-Bench 2.1 · BenchLMAgentic87.9%12 / 86
Software Engineering Benchmark VerifiedCoding80.6%16 / 114
Critical Physics TasksReasoning18.0%35 / 214
GDPval-AA normalizedAgentic54.5%23 / 133
Artificial Analysis Long Context ReasoningReasoning80.3%40 / 217
Artificial Analysis Humanity's Last ExamKnowledge41.0%43 / 222
Vibe Code Bench v1.1Coding82.3% ±3.125 / 123
Graduate-Level Google-Proof Q&AKnowledge90.1%29 / 141
LiveBench MathematicsMath95.1%14 / 66
Mystery Game PuzzlesReasoning43.0% ±5.017 / 80
Terminal-Bench 2.0 · BenchLMAgentic67.9%9 / 41
LiveBench Data AnalysisReasoning79.2%19 / 80
Humanity's Last Exam with toolsAgentic60.0%9 / 36
MMLU ProKnowledge87.0% ±0.335 / 135
ARC-AGI-1 (semi-private)Reasoning90.0%36 / 126
SimpleQA VerifiedKnowledge52.9% ±1.626 / 91
Chess PuzzlesReasoning46.0% ±5.026 / 90
ARC-AGI-2 (semi-private)Reasoning61.3%38 / 131
IMOAnswerBenchMath89.8%3 / 10
OpenHarmony Bench v1.0Coding59.0%4 / 13
Toolathlon-VerifiedAgentic74.1%8 / 23
GPQA Diamond · BenchLMKnowledge90.1%30 / 84
LiveBench LanguageKnowledge82.1%24 / 66
Code MigrationCoding41.5% ±4.328 / 77
BrowseCompAgentic83.4%23 / 63
Humanity's Last ExamKnowledge42.7%30 / 77
ToolathlonAgentic51.8%9 / 22
CyberGymAgentic83.3%12 / 29
Artificial Analysis SciCodeCoding51.0%48 / 114
FrontierMath-Tiers-1-3-v2-PrivateMath64.6% ±2.839 / 90
LiveBench ReasoningReasoning85.8%35 / 80
LiveBench Agentic CodingAgentic54.9%29 / 66
SkillsBenchCoding53.8% ±4.516 / 35
LiveBench Instruction FollowingInstruction67.7%34 / 66
MCP AtlasAgentic73.6%27 / 52
FrontierMath-Tier-4-v2-PrivateMath26.8% ±7.039 / 72
IOICoding51.6% ±2.425 / 45
LiveBench CodingCoding77.2%37 / 66
APEX-Agents-AAAgentic24.3%18 / 30
ProofBench v1.1Math50.0% ±5.033 / 54
Vibe Code Bench 1-100Coding17.5% ±3.416 / 26
Terminal-Bench 4.0Agentic14.1% ±2.039 / 61
DeepSWEAgentic62.7%38 / 59
SWE-bench ProCoding55.4%54 / 83
MirrorCodeCoding33.3% ±10.026 / 39
Terminal-Bench 2.1 · Vals AIAgentic54.7% ±1.551 / 75
Artificial Analysis Tau3-BankingAgentic39.6%15 / 22
AutomationBenchAgentic31.8%25 / 32
Agents' Last ExamAgentic25.7%19 / 21
CorpFin v2Finance65.4% ±0.927 / 131
CyberBench v1.1Other72.9% ±5.514 / 48
ARC-AGI-1 (public eval)Reasoning93.3%39 / 116
ARC-AGI-2 (public eval)Reasoning59.4%42 / 122
TaxEval v2Finance73.1% ±0.952 / 143
Legal Research BenchLegal40.9% ±3.433 / 90
MedCodeHealthcare42.5% ±2.250 / 106
LegalBenchLegal82.4% ±0.472 / 149
Tax Agent BenchFinance26.8% ±3.140 / 82
ProgramBenchCoding0.0%34 / 65
MedScribeHealthcare80.2% ±2.059 / 109
HLAB-Vals-1.1Legal10.8% ±2.719 / 35
Finance Agent (v2)Finance50.4% ±0.244 / 79
Terminal-Bench-ScienceScience4.3% ±2.426 / 44
SRE BenchOther1.9% ±0.825 / 36
EMBFinance52.8% ±3.153 / 75
LiveCodeBench Pass@1 with Chain-of-ThoughtCoding93.5%—
Apex ShortlistMath90.2%—
Chinese-SimpleQAKnowledge84.4%—
MRCR 1MReasoning83.5%—
DeepSeek DSBench FullStackCoding71.1%—
Finch (FinWorkBench): Finance & Accounting across Spreadsheet-Centric Enterprise WorkflowsAgentic67.4%—
DeepSeek DSBench HardCoding67.2%—
CorpusQA 1MReasoning62.0%—
Artificial Analysis EnterpriseOps-GymAgentic49.6%—
ApexMath38.3%—

Sources

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

History

5 releases
Index66.9 → 65.5
23 Oct 202612 Nov 2026
Output / 1M$1.98 → $4.44
23 Oct 202612 Nov 2026

Questions

How much does DeepSeek V4 Pro 0813 cost?
$0.111 per 1M input tokens, $4.44 per 1M output tokens.
How good is DeepSeek V4 Pro 0813?
Its index is 65.5 out of 100. That is place 41 of 283 ranked models.
What is DeepSeek V4 Pro 0813 best at?
Reasoning: 68.5 out of 100, place 26 of 227.
How fast is DeepSeek V4 Pro 0813?
42 output tokens per second, 0.4 s to the first token.
What is the context window of DeepSeek V4 Pro 0813?
1.05M tokens.
Is DeepSeek V4 Pro 0813 open source?
Yes. Its weights are published.
When was DeepSeek V4 Pro 0813 released?
August 2026.

Badge

DeepSeek V4 Pro 0813 on Inference360: place 41 of 283, index 65.5
HTML
<a href="https://inference360.ai/models/deepseek-v4-pro-0813"><img src="https://inference360.ai/badges/deepseek-v4-pro-0813.svg" alt="DeepSeek V4 Pro 0813 on Inference360: place 41 of 283, index 65.5" height="20"></a>
Markdown
[![DeepSeek V4 Pro 0813 on Inference360: place 41 of 283, index 65.5](https://inference360.ai/badges/deepseek-v4-pro-0813.svg)](https://inference360.ai/models/deepseek-v4-pro-0813)

Compare with

GPT-5.6 Luna65.5GPT-6 Luna65.5Qwen3.8 Max65.6Muse Spark Open65.1dots3-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 Vision Exp60.4DeepSeek V4 Flash LatestUnrankedDeepSeek V4 Flash 073161.9DeepSeek Flash LatestUnrankedDeepSeek Pro LatestUnranked