DeepSeek V4 Pro

  • Open weights
  • Reasoning
  • Input: text
  • Released 2026-08
RankedEnough results for an index.
IndexAll categories, out of 100.60.3±3.8Median 50.7
CoverageIndex weight that has results.85%
SpeedOutput tokens per second.96/sMedian 76/s
First tokenSeconds to the first output token.0.9 sMedian 1.1 s
Input / 1MUS dollars per 1M input tokens.$0.90Median $0.60
Output / 1MUS dollars per 1M output tokens.$1.50Median $2.50
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1451±4 #53Median 1425

Details

From OpenRouter
API IDThe model ID on OpenRouter.
deepseek/deepseek-v4-pro
Max outputMaximum output tokens per request.
393K
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.
$0.20
Weights
Hugging Face (opens in a new tab) ↗
RetiresDate OpenRouter removes the model.
—

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.18 counted61.462 / 221
CodingCode writing and repair.11 counted60.068 / 214
ReasoningLogic problems and puzzles.5 counted58.673 / 225
MultimodalTasks with images and text.None counted——
KnowledgeFacts and expert knowledge.6 counted62.456 / 266
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.2 counted56.020 / 63
MathMath problems.8 counted56.759 / 171

Results

50 counted
BenchmarkResultThe score from the source.PlacePlace among results on this benchmark.
Artificial Analysis GPQA DiamondKnowledge92.8%14 / 193
Artificial Analysis IFBenchInstruction76.5%11 / 140
NL2RepoCoding61.5%3 / 36
LiveCodeBenchCoding87.5% ±1.013 / 142
SWE-Bench verified · Epoch AICoding77.6% ±1.95 / 32
Critical Physics TasksReasoning18.0%34 / 207
GDPval-AA normalizedAgentic54.5%23 / 132
OTIS Mock AIME 2024-2025Math96.7% ±2.039 / 203
Terminal-Bench 2.0 · Vals AIAgentic56.2% ±5.314 / 66
Graduate-Level Google-Proof Q&AKnowledge90.1%31 / 144
GPQA diamond · Epoch AIKnowledge89.6% ±1.748 / 220
GPQA Diamond · Vals AIKnowledge89.4% ±1.730 / 135
MMLU ProKnowledge87.2% ±0.332 / 135
IOI v1Coding35.8% ±7.115 / 61
IMOAnswerBenchMath89.8%3 / 10
OpenHarmony Bench v1.0Coding59.0%4 / 13
Toolathlon-VerifiedAgentic74.1%8 / 23
BrowseCompAgentic83.4%23 / 63
GPQA Diamond · BenchLMKnowledge90.1%32 / 87
SWE-benchCoding77.4% ±1.935 / 87
ToolathlonAgentic51.8%9 / 22
CyberGymAgentic83.3%12 / 29
Humanity's Last ExamKnowledge42.7%32 / 77
Artificial Analysis SciCodeCoding51.0%48 / 113
SimpleQA VerifiedKnowledge47.0% ±1.640 / 94
LiveBench MathematicsMath90.7%31 / 66
LiveBench Data AnalysisReasoning74.5%41 / 80
MCP AtlasAgentic73.6%28 / 53
LiveBench LanguageKnowledge78.1%37 / 66
Vibe Code Bench v1.1Coding49.9% ±4.873 / 130
SkillsBenchCoding51.3% ±4.621 / 36
Code MigrationCoding26.2% ±4.046 / 77
ApexMath38.3%6 / 10
LiveBench ReasoningReasoning82.7%51 / 80
DeepSWEAgentic62.7%41 / 64
SWE-bench ProCoding55.4%54 / 83
Mystery Game PuzzlesReasoning17.0% ±3.854 / 83
APEX-Agents-AAAgentic24.3%19 / 29
FrontierMath-Tiers-1-3-v2-PrivateMath45.3% ±3.061 / 93
Terminal-Bench 4.0Agentic11.1% ±1.350 / 68
Chess PuzzlesReasoning19.0% ±3.971 / 93
Terminal-Bench 2.1 · Vals AIAgentic50.2% ±1.558 / 75
LiveBench Instruction FollowingInstruction62.4%52 / 66
AutomationBenchAgentic31.8%26 / 33
ProofBench v1.1Math16.0% ±3.747 / 55
LiveBench Agentic CodingAgentic42.6%57 / 66
LiveBench CodingCoding70.0%59 / 66
Agents' Last ExamAgentic25.7%19 / 21
FrontierMath-Tier-4-v2-PrivateMath2.4% ±2.468 / 75
Finance Agent v1.1Finance60.4% ±2.84 / 51
CorpFin v2Finance61.4% ±1.052 / 131
TaxEval v2Finance72.1% ±0.968 / 143
CaseLaw v2Legal59.4% ±1.527 / 53
Public Benefits Bench v1.1Public benefits62.9% ±1.326 / 50
Tax Agent BenchFinance25.2% ±3.147 / 89
ProgramBenchCoding0.0%35 / 66
Public Benefits Bench v1Public benefits57.6% ±1.37 / 13
LegalBenchLegal80.3% ±0.587 / 149
MedCodeHealthcare40.5% ±2.165 / 108
EMBFinance51.6% ±3.155 / 75
Finance Agent (v2)Finance44.1% ±0.758 / 79
Legal Research BenchLegal23.1% ±2.973 / 98
MedScribeHealthcare75.1% ±2.083 / 111
Apex ShortlistMath90.2%—
Chinese-SimpleQAKnowledge84.4%—
MRCR 1MReasoning83.5%—
DeepSeek DSBench HardCoding67.2%—
CorpusQA 1MReasoning62.0%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYLiveBenchNo licence statedVals AINo licence statedmodels.devMITCheaper InferenceNo licence stated

History

13 releases
Index60.7 → 60.3
18 Nov 202621 Nov 2026
Output / 1M$1.50 → $1.50
18 Nov 202621 Nov 2026

Questions

How much does DeepSeek V4 Pro cost?
$0.90 per 1M input tokens, $1.50 per 1M output tokens.
How good is DeepSeek V4 Pro?
Its index is 60.3 out of 100. That is place 72 of 282 ranked models.
What is DeepSeek V4 Pro best at?
Knowledge: 62.4 out of 100, place 56 of 266.
How fast is DeepSeek V4 Pro?
96 output tokens per second, 0.9 s to the first token.
What is the context window of DeepSeek V4 Pro?
1.05M tokens.
Is DeepSeek V4 Pro open source?
Yes. Its weights are published.
When was DeepSeek V4 Pro released?
August 2026.

Badge

DeepSeek V4 Pro on Inference360: place 72 of 282, index 60.3

Compare with

Mistral Large 460.1DeepSeek V4 Flash 073160.6GLM-5.259.9Gemini 3.1 Pro Preview60.7DeepSeek V4 Flash Vision Exp59.6

Lower price

dots3-note Preview62.3 · FreeGLM-5.3-Flash62.4 · $0.078/1MHy4 flash64.5 · $0.14/1M

More from DeepSeek

All models
DeepSeek V4 Pro 081364.0DeepSeek V4 Flash Vision Exp59.6DeepSeek V4 Flash LatestUnrankedDeepSeek V4 Flash 073160.6DeepSeek Flash LatestUnranked