DeepSeek Open weights: deepseek-ai/DeepSeek-V4-ProReasoning modelAuto-detected

DeepSeek V4 Pro

Name: DeepSeek V4 Pro benchmark results
Creator: BenchLeader

Best configuration ranks #95 of 610 on the BenchLeader Index at 60.0 ±3.0 (high reasoning effort). Last measured 8 Sept 2026. Released 24 Apr 2026.

Blended price: $0.548/M; $0.440 in · $0.870 out
Output speed: 66 tok/s
First answer: 32 s; first token 1.39 s
Context: 1M

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

Overall index60
Reasoning62
Agents & tools64
Knowledge59
Instruction following69
Long context62
Composite67

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. “Default” means the publisher did not say which setting was used.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
no reasoning	48.8	#316	70 tok/s	1.75 s	$0.0004	Agents & tools 65 · Coding 32 · Composite 55 · Instruction following 47 · Knowledge 50 · Long context 52 · Maths 45 · Reasoning 41
low	–	–	79 tok/s	27 s	$0.0004	Reasoning 61
highbest	60.0	#95	66 tok/s	32 s	$0.0004	Agents & tools 64 · Composite 67 · Instruction following 69 · Knowledge 59 · Long context 62 · Reasoning 62
max	56.1	#163	79 tok/s	27 s	$0.0004	Agents & tools 44 · Coding 61 · Knowledge 59 · Maths 57 · Reasoning 64
default	59.7	#98	79 tok/s	27 s	$0.0004	Agents & tools 53 · Coding 56 · Composite 67 · Human preference 59 · Instruction following 74 · Knowledge 64 · Long context 67 · Maths 58 · Reasoning 58

Benchmark results

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.

Reasoning

Benchmark	no reasoning	low	high	max	default	Source
GPQA Diamond	73.2%#145	–	–	91.7%#24	–	Epoch AI Benchmarking Hub
LMArena Hard Prompts	–	–	–	–	1480#53	LMArena
LiveBench Reasoningnot in index	–	–	–	–	85.8%#25	LiveBench
GPQA Diamond (AA)not in index	71.7%#274	–	90.5%#59	–	92.8%#28	Artificial Analysis
Humanity's Last Exam (AA)not in index	8.3%#280	–	35.2%#79	–	41.0%#50	Artificial Analysis
GPQA Diamond (Vals)not in index	–	–	–	92.4%#16	–	Vals AI
Kagi LLM Benchmark	–	–	–	–	53.5%#70	Kagi LLM Benchmark
ARC-AGI-1	13.0%#165	90.5%#50	87.2%#61	90.0%#53	–	ARC Prize
ARC-AGI-2	0.8%#164	56.3%#63	59.7%#59	61.3%#54	–	ARC Prize

Coding

Benchmark	no reasoning	low	high	max	default	Source
SWE-bench Verified (Epoch)	–	–	–	77.6%#6	–	Epoch AI Benchmarking Hub
SciCode	–	–	–	50.0%#60	–	SciCode
WeirdML	–	–	–	66.2%#36	–	WeirdML
FrontierCode	17.6%#25	–	–	–	–	Cognition
LMArena Coding	–	–	–	–	1502#61	LMArena
LMArena WebDev	–	–	–	–	1446#55	LMArena
LiveBench Codingnot in index	–	–	–	–	77.2%#28	LiveBench
SciCode (AA)not in index	–	–	–	–	51.0%#62	Artificial Analysis
LiveCodeBench	–	–	–	87.5%#13	–	Vals AI
IOI	–	–	–	35.8%#15	–	Vals AI
SWE-bench (Vals)not in index	–	–	–	96.4%#2	–	Vals AI

Agents & tools

Benchmark	no reasoning	low	high	max	default	Source
LMArena Agent	–	–	4.4#12	–	-0.4#26	LMArena
LiveBench Agentic Codingnot in index	–	–	–	–	55.0%#22	LiveBench
Terminal-Bench Hard	36.4%#69	–	41.7%#48	–	46.2%#30	Artificial Analysis
τ²-Bench Telecom (AA)not in index	91.2%#59	–	94.2%#34	–	96.2%#15	Artificial Analysis
Terminal-Bench 2.1 (Vals)	–	–	–	54.7%#40	50.2%#48	Vals AI

Maths

Benchmark	no reasoning	low	high	max	default	Source
FrontierMath Tiers 1–3	–	–	–	64.6%#24	–	Epoch AI Benchmarking Hub
FrontierMath Tier 4	–	–	–	26.8%#31	–	Epoch AI Benchmarking Hub
OTIS Mock AIME	46.7%#169	–	–	98.6%#13	–	Epoch AI Benchmarking Hub
ProofBench	–	–	–	16.0%#41	50.0%#17	Vals AI
LiveBench Mathematicsnot in index	–	–	–	–	95.1%#8	LiveBench
AIME 2026	–	–	–	96.7%#6	–	MathArena
HMMT February 2026	–	–	–	93.9%#10	–	MathArena
MathArena Apex	–	–	–	28.1%#10	–	MathArena

Knowledge

Benchmark	no reasoning	low	high	max	default	Source
SimpleQA Verified	–	–	–	52.9%#17	–	Epoch AI Benchmarking Hub
LiveBench Data Analysisnot in index	–	–	–	–	79.2%#11	LiveBench
AA-Omniscience	-29.9#214	–	-10.6#144	–	0.8#89	Artificial Analysis
MMLU-Pro	–	–	–	87.3%#32	–	Vals AI
LegalBench	–	–	–	82.4%#65	–	Vals AI
CorpFin	–	–	–	65.4%#28	–	Vals AI
TaxEval	–	–	–	73.1%#53	–	Vals AI

Instruction following

Benchmark	no reasoning	low	high	max	default	Source	Trend
LiveBench Languagenot in index	–	–	–	–	82.1%#19	LiveBench
IFBench	45.8%#197	–	71.3%#57	–	76.5%#20	Artificial Analysis

Human preference

Benchmark	no reasoning	low	high	max	default	Source	Trend
LMArena Text	–	–	–	–	1458#54	LMArena
EQ-Bench 4	–	–	–	–	1166#17	EQ-Bench

Long context

Benchmark	no reasoning	low	high	max	default	Source	Trend
AA-LCR	53.0%#243	–	70.3%#157	–	80.3%#45	Artificial Analysis

Composite

Benchmark	no reasoning	low	high	max	default	Source
Epoch Capabilities Indexnot in index	–	–	–	–	155.4#21	Epoch AI Benchmarking Hub
LiveBench	–	–	–	–	77.4%#15	LiveBench
AA Intelligence Index	20.8#165	–	30.1#81	–	36.3#50	Artificial Analysis
Vals Indexnot in index	–	–	–	52.4#25	–	Vals AI

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
NovitaAI	67 tok/s	1.39 s	$1.60	$3.20	1.0M	fp8
Baseten	62 tok/s	0.60 s	$1.74	$3.48	1.0M	fp4
Parasail	62 tok/s	0.87 s	$1.74	$3.48	1.0M	fp8
Baseten (US)	62 tok/s	0.58 s	$1.74	$3.48	1.0M	fp4
Baidu Qianfan	54 tok/s	1.15 s	$0.869	$1.74	1.0M	fp8
Azure (US)	47 tok/s	1.40 s	$1.91	$3.83	1.0M	–
NextBit	46 tok/s	2.15 s	$1.80	$3.55	1.0M	fp8
GMICloud	41 tok/s	2.13 s	$0.957	$1.91	1.0M	fp8
StreamLake	39 tok/s	1.54 s	$0.870	$1.74	1.0M	fp8
SiliconFlow	37 tok/s	2.15 s	$1.50	$3.13	1.0M	fp8
AtlasCloud	37 tok/s	1.31 s	$1.68	$3.38	1.0M	fp4
DigitalOcean	35 tok/s	0.80 s	$0.870	$1.74	1.0M	–
Alibaba Cloud Int.	33 tok/s	1.66 s	$1.42	$2.83	1M	fp8
Venice	28 tok/s	1.75 s	$1.65	$3.30	1M	–
DeepInfra	27 tok/s	1.31 s	$1.30	$2.60	1.0M	fp8

Price history

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

input outputnow $0.870 in · $1.74 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 $0.004 per 1M. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0004	$0.0003	36.6 s
Summarise a 30-page report	12,000 / 600	$0.0058	$0.0019	41.2 s
Code edit	6,000 / 1,500	$0.0039	$0.0020	54.9 s
Agentic coding session	60,000 / 4,000	$0.030	$0.010	1.5 min
Structured extraction	2,000 / 200	$0.0011	$0.0004	35.1 s