DeepSeek Open weights: deepseek-ai/DeepSeek-V4.1-FlashReasoning modelCompare with another model

DeepSeek V4.1 Flash

Reasoning effort

DeepSeek V4.1 Flash is a DeepSeek open-weights reasoning model, released 10 Sept 2026. Its best configuration (max reasoning effort) ranks #90 of 775 on the BenchLeader Index at 62.0 ±3.5, in the upper half. The ± is the point: 174 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in composite (74) and lowest in knowledge (55). At $0.525 per million tokens blended it is mid-priced. Output speed of 240 tokens per second puts it in the fastest quarter, with a first answer in 9.3 s. It has been superseded by DeepSeek V4.1 Flash 1017, which is the version to compare against today. It has been measured at 5 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 19 Oct 2026.

Blended price: $0.525/M; $0.300 in · $1.20 out
Output speed: 240 tok/s; measured by Artificial Analysis
First answer: 9.25 s; first token 0.70 s
Context: 1.0M
Full answer: 11 s; median, reasoning included
Cost per run: $0.571; one full Intelligence Index run
Released: 10 Sept 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index62
Reasoning63
Coding66
Agents & tools58
Maths62
Knowledge55
Human preference67
Multimodal61
Long context67
Composite74

Versions

DeepSeek has shipped 2 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.

Model	Released	Index	Rank
NewDeepSeek V4.1 Flash 1017max	17 Oct 2026	63.1	#72
DeepSeek V4.1 Flashmaxthis page	10 Sept 2026	62.0	#90

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	48.0	#435	442 tok/s	0.82 s	$0.0005	Coding 34 · Composite 57 · Knowledge 57 · Long context 52 · Multimodal 48 · Reasoning 36
low	–	–	98 tok/s	0.70 s	$0.0002	Reasoning 56
high	54.3	#252	98 tok/s	0.70 s	$0.0002	Agents & tools 61 · Coding 48 · Knowledge 58 · Reasoning 57
maxbest	62.0	#90	240 tok/s	9.25 s	$0.0005	Agents & tools 58 · Coding 66 · Composite 74 · Human preference 67 · Knowledge 55 · Long context 67 · Maths 62 · Multimodal 61 · Reasoning 63
not stated	–	–	98 tok/s	0.70 s	$0.0002	Agents & tools 39 · Composite 61 · Maths 56

What more thinking costs

Turning the effort up buys index points and multiplies the bill. Here max costs 2.3× no reasoning for +14.0 on the index — 11.7 points per doubling of spend. Compare that with other models.

How its index has moved

19 Sept 2026 to 20 Oct 2026

max
high
no reasoning

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Metadata

Benchmark	max	Source	Trend
AA Openness Indexnot in index	44.4#27	Artificial Analysis

Reasoning

Benchmark	no reasoning	low	high	max	not stated	Source
GPQA Diamond	–	–	–	89.8%#49	–	Epoch AI Benchmarking Hub
Arena Hard Prompts	–	–	–	1499#36	–	Arena (LMArena)
LiveBench 2.0 Reasoningnot in indexeffort not stated	–	–	–	–	86.7%#26	LiveBench
LiveBench 1.0 Reasoningnot in index	–	–	–	86.7%#31	–	LiveBench
Humanity's Last Exam (AA)not in index	10.8%#307	–	–	39.3%#91	–	Artificial Analysis
ARC-AGI-1	7.8%#223	90.5%#68	88.5%#81	94.5%#37	–	ARC Prize
ARC-AGI-2	1.3%#204	60.6%#80	67.5%#61	72.9%#50	–	ARC Prize
ARC-AGI-3	0.5%#42	3.3%#28	3.2%#29	2.7%#31	–	ARC Prize
CritPt	0.3%#266	–	–	14.3%#97	–	Artificial Analysis
MysteryMechanismnot in index	–	–	21.2%#21	–	–	Vals AI
Chess Puzzlesnot in index	–	–	–	48.0%#19	–	Epoch AI Benchmarking Hub
Mystery Game Puzzlesnot in index	–	–	–	43.0%#15	–	Epoch AI Benchmarking Hub
LMCAnot in indexeffort not stated	–	–	–	–	47.0%#82	Epoch AI Benchmarking Hub
DTBenchnot in indexeffort not stated	–	–	–	–	89.9%#88	Epoch AI Benchmarking Hub

Coding

Benchmark	no reasoning	high	max	not stated	Source
SciCode	35.5%#190	–	51.9%#90	–	SciCode
FrontierCode	–	43.9%#17	–	–	Cognition
Arena Coding	–	–	1523#29	–	Arena (LMArena)
Arena WebDev	–	–	1664#12	–	Arena (LMArena)
LiveBench 2.0 Codingnot in indexeffort not stated	–	–	–	90.2%#12	LiveBench
LiveBench 1.0 Codingnot in index	–	–	80.0%#22	–	LiveBench
SciCode (AA)not in index	35.5%#198	–	51.9%#89	–	Artificial Analysis
IOI	–	40.3%#34	–	–	Vals AI
Code Migrationnot in index	–	45.6%#16	–	–	Vals AI
ProgramBenchnot in index	–	–	0.5%#18	–	Vals AI
Vibe Code Bench 1-100not in index	–	16.4%#16	–	–	Vals AI
Vibe Code Bench v1.1not in index	–	84.7%#17	–	–	Vals AI
Surface Evolver Benchnot in index	–	46.3%#20	–	–	Epoch AI Benchmarking Hub
ALE-Benchnot in index	–	–	1092.3#52	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	5.6%#98	–	26.8%#48	–	Artificial Analysis
Startrise Frontendnot in indexeffort not stated	–	–	–	75.6%#9	Startrise
FrontierCode 1.1 (AA)not in index	–	–	44.0%#13	–	Artificial Analysis

Agents & tools

Benchmark	no reasoning	high	max	not stated	Source
APEX-Agentseffort not stated	–	–	–	39.5%#43	Mercor
Arena Agent	–	–	2.6#19	–	Arena (LMArena)
LiveBench Long Agenticnot in indexeffort not stated	–	–	–	40.1%#24	LiveBench
LiveBench Browsernot in indexeffort not stated	–	–	–	84.0%#10	LiveBench
LiveBench Agentic Codingnot in index	–	–	77.3%#1	–	LiveBench
Terminal-Bench 2.1 (Vals)	–	74.5%#18	–	–	Vals AI
GDPval-AA v2.1not in index	42.1%#91	–	55.0%#33	–	Artificial Analysis
ITBench SRE (AA)not in index	–	–	46.9%#10	–	Artificial Analysis
Analyst Agent (AA)not in index	–	–	35.0%#28	–	Artificial Analysis
CyberBenchnot in index	–	73.7%#10	–	–	Vals AI
Finance Agent v2not in index	–	53.5%#33	–	–	Vals AI
Legal Research Benchnot in index	–	–	41.4%#25	–	Vals AI
SkillsBenchnot in index	–	69.8%#1	–	–	Vals AI
SREBenchnot in index	–	–	0.8%#26	–	Vals AI
Tax Agent Benchnot in index	–	–	26.6%#34	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	–	19.7%#28	–	–	Vals AI
Terminal-Bench Sciencenot in index	–	0.0%#37	–	–	Vals AI
AutomationBenchnot in index	47.8%#83	–	68.9%#6	–	Artificial Analysis
GDP.pdf (AA)not in index	3.8%#170	–	12.8%#107	–	Artificial Analysis
Harvey LAB-AA v1.1not in index	–	–	1.7%#16	–	Harvey
AA-Briefcase v1.1not in index	–	–	1419#40	–	Artificial Analysis
Terminal-Bench-Science 0.1 (AA)not in index	–	–	9.1%#30	–	Artificial Analysis
CWE-Bench (AA)	–	–	68.3%#1	–	Artificial Analysis
DeepSec-Bench (AA)	–	–	31.2%#3	–	Artificial Analysis
CyberGym end-to-end (AA)	–	–	22.9%#11	–	Artificial Analysis

Maths

Benchmark	max	not stated	Source
FrontierMath Tiers 1–3	67.4%#32	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	26.8%#42	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	98.3%#23	–	Epoch AI Benchmarking Hub
ProofBencheffort not stated	–	54.0%#27	Vals AI
LiveBench Mathematicsnot in index	93.3%#23	–	LiveBench
Arena Maths	1495#22	–	Arena (LMArena)

Knowledge

Benchmark	no reasoning	high	max	not stated	Source
SimpleQA Verified	–	–	42.9%#42	–	Epoch AI Benchmarking Hub
LiveBench 2.0 Data Analysisnot in indexeffort not stated	–	–	–	79.3%#14	LiveBench
LiveBench Researchnot in indexeffort not stated	–	–	–	86.8%#4	LiveBench
LiveBench 1.0 Data Analysisnot in index	–	–	79.3%#17	–	LiveBench
AA-Omniscience	-10.2#191	–	-5.3#164	–	Artificial Analysis
LegalBench	–	83.3%#55	–	–	Vals AI
BioMysteryBenchnot in index	–	67.8%#15	–	–	Vals AI
Excel Modeling Benchmarknot in index	–	57.2%#39	–	–	Vals AI
MedCodenot in index	–	41.2%#55	–	–	Vals AI
MedScribenot in index	–	85.5%#27	–	–	Vals AI
Public Benefits Benchnot in index	–	–	64.3%#22	–	Vals AI
GDP.pdfnot in index	–	–	19.8%#31	–	Epoch AI Benchmarking Hub
AA-Omniscience: accuracynot in index	28.4%#192	–	46.4%#80	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	46.0%#138	–	3.5%#537	–	Artificial Analysis

Instruction following

Benchmark	max	Source
LiveBench Languagenot in index	81.2%#28	LiveBench
LiveBench Instruction Followingnot in index	70.0%#30	LiveBench
Arena Instruction Followingnot in index	1476#24	Arena (LMArena)

Human preference

Benchmark	max	Source
Arena Text	1475#38	Arena (LMArena)
Arena Creative Writingnot in index	1440#58	Arena (LMArena)
Arena Multi-turnnot in index	1468#65	Arena (LMArena)
Arena Longer Queriesnot in index	1481#42	Arena (LMArena)

Multimodal

Benchmark	no reasoning	high	max	not stated	Source
Arena Vision	–	–	1278#38	–	Arena (LMArena)
LiveBench Designnot in indexeffort not stated	–	–	–	79.2%#15	LiveBench
MMMU-Pro	65.6%#172	–	77.0%#84	–	Artificial Analysis
SAGEnot in index	–	47.9%#29	–	–	Vals AI

Long context

Benchmark	no reasoning	max	Source	Trend
AA-LCR	55.3%#295	84.0%#11	Artificial Analysis
MLCRnot in index	–	22.8%#24	Artificial Analysis

Composite

Benchmark	no reasoning	max	not stated	Source
Epoch Capabilities Indexnot in indexeffort not stated	–	–	155.9#32	Epoch AI Benchmarking Hub
LiveBench 2.0effort not stated	–	–	78.0%#12	LiveBench
LiveBench 1.0not in index	–	81.1%#9	–	LiveBench
AA Intelligence Index v4.3.2	24.7#171	39.5#68	–	Artificial Analysis
AA Cyber Indexnot in index	–	40.8#8	–	Artificial Analysis

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

Provider	Speed	First token	Input $/M	Output $/M	Context	Quantisation
Fireworks Spot	259 tok/s	0.38 s	$0.210	$0.840	1.0M	–
Fireworks	259 tok/s	0.38 s	$0.450	$1.80	1.0M	–
Baseten	230 tok/s	0.36 s	$0.300	$1.20	1.0M	–
Together	197 tok/s	0.32 s	$0.300	$1.20	1.0M	–
Baseten	195 tok/s	0.21 s	$0.300	$1.20	1.0M	–
Baidu Qianfan	188 tok/s	0.55 s	$0.300	$1.20	1.0M	fp8
DigitalOcean	171 tok/s	0.68 s	$0.225	$0.900	1.0M	–
CoreWeave	167 tok/s	0.76 s	$0.150	$0.650	1.0M	fp8
Prime Intellect	156 tok/s	0.43 s	$0.300	$1.20	1.0M	–
Baseten	150 tok/s	0.29 s	$0.600	$2.40	1.0M	–
Parasail	136 tok/s	0.55 s	$0.300	$1.20	1.0M	fp8
Venice	136 tok/s	0.40 s	$0.300	$1.20	1M	fp8
NovitaAI	135 tok/s	1.60 s	$0.195	$0.780	1.0M	fp8
inference.net	132 tok/s	0.38 s	$0.103	$0.715	1.0M	fp8
Morph	124 tok/s	0.64 s	$0.172	$0.600	1.0M	fp8
inference.net	121 tok/s	0.28 s	$0.103	$0.715	1.0M	fp8
Modal	114 tok/s	0.39 s	$0.300	$1.20	1.0M	–
Wafer	114 tok/s	0.41 s	$0.190	$0.650	1.0M	mxfp4
Alibaba Cloud Int.	108 tok/s	0.98 s	$0.330	$1.32	1M	–
Makora	107 tok/s	0.43 s	$0.260	$1.17	1.0M	fp8
AtlasCloud	102 tok/s	0.86 s	$0.135	$0.540	1.0M	fp8
Ionstream	92 tok/s	1.15 s	$0.270	$0.750	1.0M	–
Decart	87 tok/s	1.56 s	$0.090	$0.180	1.0M	fp4
Fireworks	82 tok/s	0.92 s	$0.300	$1.20	1.0M	–
Phala	81 tok/s	2.34 s	$0.210	$0.840	1.0M	–
Google Vertex	79 tok/s	0.67 s	$0.300	$1.20	1.0M	–
SiliconFlow	74 tok/s	1.63 s	$0.300	$1.20	1.0M	fp8
GMICloud	67 tok/s	2.35 s	$0.180	$0.720	1.0M	fp8
Nebius Token Factory	59 tok/s	0.69 s	$0.300	$1.20	1.0M	fp8
DeepInfra	58 tok/s	0.79 s	$0.140	$0.420	1.0M	fp8
StreamLake	57 tok/s	1.31 s	$0.147	$0.588	1.0M	fp8
Open Inference	54 tok/s	0.84 s	$0.056	$0.120	1.0M	fp4
DeepInfra	49 tok/s	0.43 s	$0.210	$0.630	1.0M	fp8
Sail Research	41 tok/s	2.64 s	$0.101	$0.150	1.0M	fp4
DeepInfra	39 tok/s	1.89 s	$0.112	$0.336	1.0M	fp8
Sail Research	34 tok/s	1.82 s	$0.101	$0.150	1.0M	fp4

Price history

Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.

input outputnow $0.200 in · $0.600 out

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.003 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0005	$0.0004	10.5 s
Summarize a 30-page report	12,000 / 600	$0.0043	$0.0016	11.8 s
Code edit	6,000 / 1,500	$0.0036	$0.0023	15.5 s
Agentic coding session	60,000 / 4,000	$0.023	$0.0094	25.9 s
Structured extraction	2,000 / 200	$0.0008	$0.0004	10.1 s