Inference
EN
DeepSeek

DeepSeek V4.1 Flash 1017

DeepSeek V4.1 Flash 1017 is a reasoning model from DeepSeek in the DeepSeek V4.1 Flash family. 61 benchmarks count toward its index, in 7 categories.

  • FastestFastest with top-25% index.
  • Open weights
  • Reasoning
  • Input: text · image
  • Released 2026-10
RankedEnough results for an index.
IndexAll categories, out of 100.68.5±2.6Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.261/sMedian 67/s
First tokenSeconds to the first output token.0.2 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$0.022batch $0.011Median $0.60
Output / 1MUS dollars per 1M output tokens.$0.40batch $0.20 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.1.05MMedian 262K
EloLMArena rating and rank.1463±6 #39Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
deepseek/deepseek-v4.1-flash
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.
<$0.01
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
OpenRouterRouter$0.022$0.401.05M
VultrCloud$0.15$0.601.05M
Deep InfraHost$0.20$0.601.05M
CoreWeaveHost$0.13$0.651.05M
Arcee AIHost$0.22$0.661M
Fireworks AIHost$0.21$0.841M
BasetenHost$0.30$1.201.05M
DigitalOceanCloud$0.30$1.201.05M
Microsoft AzureCloud$0.30$1.201M
ModalHost$0.30$1.201M
Novita AIHost$0.30$1.201.05M
Together AIHost$0.30$1.201.05M
TinfoilHost$0.60$1.451.05M

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

Run locally

763.21B parameters
16-bit
1603.7 GB
8-bit
852.5 GB
4-bit
486.8 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.20 counted72.214 / 218
CodingCode writing and repair.16 counted69.821 / 209
ReasoningLogic problems and puzzles.11 counted65.045 / 227
MultimodalTasks with images and text.3 counted62.236 / 84
KnowledgeFacts and expert knowledge.5 counted64.247 / 268
MultilingualTasks in many languages.None counted——
InstructionTasks with strict prompt rules.1 counted68.3—
MathMath problems.5 counted65.428 / 168

Results

61 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
LiveBench Agentic CodingAgenticAbove fitFar above this model's index.77.3%1 / 66
NL2RepoCoding68.2%1 / 36
SkillsBenchCoding69.8% ±3.91 / 35
Terminal-Bench 2.1 · BenchLMAgentic90.6%3 / 86
Artificial Analysis Long Context ReasoningReasoning84.0%8 / 217
Artificial Analysis CWE-Bench-AAAgentic68.3%1 / 19
DeepSWEAgentic77.0%4 / 59
LiveBench ResearchAgentic86.8%4 / 52
AutomationBenchAgenticAbove fitFar above this model's index.60.5%3 / 32
CyberGymAgentic88.1%3 / 29
Humanity's Last Exam with toolsAgentic63.9%5 / 36
OTIS Mock AIME 2024-2025Math98.3% ±0.628 / 200
Agent Arena command recoveryAgentic7.97 / 50
Vibe Code Bench v1.1Coding84.7% ±2.918 / 123
GDPval-AA normalizedAgentic55.0%21 / 133
ARC-AGI-1 (semi-private)Reasoning94.5%21 / 126
Graduate-Level Google-Proof Q&AKnowledge90.9%25 / 141
LiveBench Data AnalysisReasoning79.3%15 / 80
Artificial Analysis DeepsecBench-AAAgentic31.2%4 / 21
ARC-AGI-2 (semi-private)Reasoning72.9%25 / 131
GPQA diamond · Epoch AIKnowledge89.8% ±1.742 / 217
Artificial Analysis Omniscience AccuracyKnowledge46.4%33 / 164
Chess PuzzlesReasoning48.0% ±5.019 / 90
Mystery Game PuzzlesReasoning43.0% ±5.017 / 80
Code MigrationCoding45.6% ±4.317 / 77
LiveBench Browser UseAgentic84.0%11 / 49
Artificial Analysis Humanity's Last ExamKnowledge39.2%51 / 222
MirrorCodeCoding64.4% ±11.49 / 39
OpenHarmony Bench v1.0Coding60.3%3 / 13
Terminal-Bench 2.1 · Vals AIAgentic74.5% ±1.618 / 75
Critical Physics TasksReasoning14.3%52 / 214
LiveBench Coding 2.0Coding90.2%13 / 52
GPQA Diamond · BenchLMKnowledge90.9%24 / 84
LiveBench DesignAgentic79.2%15 / 52
LiveBench MathematicsMath93.3%20 / 66
LiveBench CodingCoding80.0%20 / 66
Agent Arena steerabilityAgentic4.018 / 54
MazeBench: visual spatial reasoning in a 3D open worldAgentic1.0%16 / 48
ScarfBench: Cross-Framework Application Migration in Enterprise JavaCoding69.0%4 / 11
FrontierMath-Tiers-1-3-v2-PrivateMath67.4% ±2.833 / 90
Artificial Analysis MMMU-ProMultimodal77.0%43 / 113
Artificial Analysis SciCodeCoding51.9%44 / 114
LiveBench ReasoningReasoning86.7%31 / 80
LiveBench LanguageKnowledge81.2%26 / 66
Bug Hunt BenchCoding39.1%10 / 25
Humanity's Last ExamKnowledge42.6%31 / 77
Artificial Analysis ITBench-AAAgentic46.9%10 / 24
cursorBench40Coding47.7%9 / 21
LiveBench Long AgenticAgentic40.1%22 / 51
LiveBench Instruction FollowingInstruction70.0%29 / 66
CWE-bench v2Agentic57.6%5 / 11
Agent Arena task outcomeAgentic2.325 / 54
Artificial Analysis Cyber IndexAgentic40.8%9 / 19
SimpleQA VerifiedKnowledge42.9% ±1.647 / 91
Agents' Last ExamAgentic35.4%11 / 21
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal64.4%21 / 39
ARC-AGI-3 (semi-private)Reasoning0.9%14 / 26
FrontierMath-Tier-4-v2-PrivateMath26.8% ±7.039 / 72
Terminal-Bench 4.0Agentic19.7% ±1.834 / 61
ProofBench v1.1Math54.0% ±5.031 / 54
ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Multimodal87.1%15 / 26
Artificial Analysis CyberGym-E2E-AAAgentic22.9%12 / 19
Furniture AssemblyReasoning34.2% ±6.024 / 37
Vibe Code Bench 1-100Coding16.4% ±3.217 / 26
jevBench15Reasoning6.6%68 / 103
CWE-bench v1Agentic55.0%8 / 12
ExploitGymAgentic15.3%16 / 23
ProgramBench: Can Language Models Rebuild Programs From Scratch?Coding29.1%14 / 19
IOICoding40.3% ±2.635 / 45
FrontierCode 2.0Coding19.3%23 / 27
jevBench14Reasoning4.8%85 / 89
ARC-AGI-1 (public eval)Reasoning98.0%11 / 116
ARC-AGI-2 (public eval)Reasoning81.7%23 / 122
CyberBench v1.1Other73.7% ±5.510 / 48
MedScribeHealthcare85.5% ±1.927 / 109
ProgramBenchCoding0.5% ±0.519 / 65
Legal Research BenchLegal41.3% ±3.430 / 90
LegalBenchLegal83.3% ±0.556 / 149
Finance Agent (v2)Finance53.5% ±0.433 / 79
Public Benefits Bench v1.1Public benefits64.3% ±1.223 / 50
Tax Agent BenchFinance26.6% ±3.141 / 82
MedCodeHealthcare41.2% ±2.056 / 106
EMBFinance57.2% ±3.240 / 75
SAGEEducation85.4% ±2.017 / 29
BioMysteryBenchScience67.8% ±1.118 / 30
SRE BenchOther0.8% ±0.528 / 36
MysteryMechanismScience21.2% ±2.724 / 30
Terminal-Bench-ScienceScience0.0%41 / 44
BabyVision with PythonMultimodal92.5%—
Chartography with image and code toolsMultimodal83.0%—
ApexMath65.6%—
ZeroBench_main with PythonMultimodal51.0%—
terminalBench3Agentic30.0%—
STEMBench120: six-domain scientific research progressReasoning2.5%—

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYARC Prize FoundationTerms restrict reuseLiveBenchNo licence statedVals AINo licence statedLMArenaCC BY 4.0models.devMIT

History

5 releases
Index69.8 → 68.5
23 Oct 202612 Nov 2026
Output / 1M$0.144 → $0.40
23 Oct 202612 Nov 2026

Questions

How much does DeepSeek V4.1 Flash 1017 cost?
$0.022 per 1M input tokens, $0.40 per 1M output tokens.
How good is DeepSeek V4.1 Flash 1017?
Its index is 68.5 out of 100. That is place 27 of 283 ranked models.
What is DeepSeek V4.1 Flash 1017 best at?
Agentic: 72.2 out of 100, place 14 of 218.
How fast is DeepSeek V4.1 Flash 1017?
261 output tokens per second, 0.2 s to the first token.
What is the context window of DeepSeek V4.1 Flash 1017?
1.05M tokens.
Is DeepSeek V4.1 Flash 1017 open source?
Yes. Its weights are published.
When was DeepSeek V4.1 Flash 1017 released?
October 2026.

Badge

DeepSeek V4.1 Flash 1017 on Inference360: place 27 of 283, index 68.5
HTML
<a href="https://inference360.ai/models/deepseek-v4-1-flash"><img src="https://inference360.ai/badges/deepseek-v4-1-flash.svg" alt="DeepSeek V4.1 Flash 1017 on Inference360: place 27 of 283, index 68.5" height="20"></a>
Markdown
[![DeepSeek V4.1 Flash 1017 on Inference360: place 27 of 283, index 68.5](https://inference360.ai/badges/deepseek-v4-1-flash.svg)](https://inference360.ai/models/deepseek-v4-1-flash)

Compare with

Gemini 3.7 Flash68.5Step 5 Preview68.4GPT-5.6 Terra68.9GPT-5.569.1Grok 4.769.1

Lower price

Ling 3.1 Flash67.2 · $0.135/1M

More from DeepSeek

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