What is the IQ of deepseek-v4-flash?

deepseek-v4-flash has an estimated IQ of 121, which ranks #42 out of 176 public models on the AI IQ leaderboard. The score combines seven equally weighted reasoning dimensions using measured benchmark results and conservative estimates for missing coverage.

What is deepseek-v4-flash best at?

Among the dimensions with published benchmark results, deepseek-v4-flash ranks highest in Mathematical Reasoning (IQ 131, #30 of 176) and Abstract Reasoning (IQ 103, #46 of 201). Its lowest-ranked measured dimension is Reliability (IQ 114, #75 of 177).

What is the EQ of deepseek-v4-flash?

deepseek-v4-flash has an estimated EQ of 113. EQ measures emotional reasoning — how well a model reads and responds to emotional context — and is scored separately from IQ.

Which benchmarks has deepseek-v4-flash been tested on?

deepseek-v4-flash has published results on 27 of the 39 benchmarks behind AI IQ's Composite IQ, including IFBench, MathArena, CritPt, GPQA Diamond and GDPval-AA v2.1 normalized score.

How much does deepseek-v4-flash cost?

deepseek-v4-flash costs $0.44 per million input tokens and $1.32 per million output tokens. Adjusted for how many tokens it actually consumes on real tasks, its effective cost works out to $1.69 per million input-plus-output tokens.

Is deepseek-v4-flash open source?

Yes — deepseek-v4-flash is an open-weights model: its trained weights are published, so it can be downloaded and self-hosted.

Who developed deepseek-v4-flash?

deepseek-v4-flash was developed by DeepSeek, based in China.

When was deepseek-v4-flash released?

deepseek-v4-flash was publicly released on April 24, 2026.

How large is the context window of deepseek-v4-flash?

deepseek-v4-flash supports a context window of 1,000,000 tokens — the amount of text it can consider in a single request.

How fast is deepseek-v4-flash?

In independent measurements, deepseek-v4-flash generates about 142 tokens per second, with a typical response time of about 19 seconds.

DeepSeekReleased Apr 24, 2026

Open weights

deepseek-v4-flash

Strongest in Mathematical Reasoning (#30 of 176) and Abstract Reasoning (#46 of 201); weakest in Reliability (#75 of 177). All seven dimensions are measured.

Composite IQ ?

121

#42 of 176 ranked models · ahead of 77%

DeepSeek's #4 model, behind DeepSeek V4.1 Flash 1017 (128)

Effective cost

$1.69/1M

$0.44 in · $1.32 out

Speed

142tok/s

19s typical response

Context

1Mtokens

1,000,000 tokens in

Evidence

7of 7 dims

27 of 39 benchmarks

Compare with gpt-6.1-solAPI pricing

IQ profile

Seven dimensions, averaged equally into Composite IQ

Compare shape with

Measured95% range

#1 in dimensionTop-10 median

IQ dimensions

Ranks are among models scored in each dimension. Ranges are model-based estimates, not calibrated confidence intervals.

DimensionIQPosition · 70–160RankEvidencevs Top 10

Abstract ReasoningAbstract

103measured

#46 / 201

4/5

-23.5

Mathematical ReasoningMath

131measured

#30 / 176

4/5

-16

Academic ReasoningAcademic

128measured

#56 / 203

4/6

-7.5

Programmatic ReasoningCoding

127measured

#47 / 177

4/6

-22.5

Computer UseComputer Use

122measured

#41 / 176

2/5

-21

ReliabilityReliability

114measured

#75 / 177

5/6

-20

Knowledge WorkKnowledge Work

122measured

#48 / 176

4/6

-9.5

deepseek-v4-flash95% range (model-based, uncalibrated)Top-10 median#1 in dimension100 = average human

Benchmark results

27 of 39 Composite IQ benchmarks have a published result for deepseek-v4-flash. Benchmarks without a result stay listed so gaps are visible.

Abstract Reasoning IQ 103 · measuredBest: #20 of 55 on Mystery Game Puzzles4/5

BenchmarkScorePercentileRankLeader · source

Mystery Game Puzzles

Full source leaderboard reconciliation vEpoch 100-position leaderboard (task implementation versions recorded per run) · Oct 23

34.0%

#20 / 55

Leader gpt-6-astra · 84.0%
Full source leaderboard reconciliation vEpoch 100-position leaderboard (task implementation versions recorded per run) · Oct 23

Chess Puzzles

Epoch AI vEpoch 100-position leaderboard (task implementation versions recorded per run) · Oct 14

33.0%

#29 / 76

Leader gpt-6-astra · 72.0%
Epoch AI vEpoch 100-position leaderboard (task implementation versions recorded per run) · Oct 14

ARC-AGI-2

Full source leaderboard reconciliation · Oct 23

61.4%

#26 / 68

Leader gpt-6-astra · 95.0%
Full source leaderboard reconciliation · Oct 23

ARC-AGI-1

Full source leaderboard reconciliation · Oct 23

89.0%

#30 / 68

Leader fable-5 · 98.5%
Full source leaderboard reconciliation · Oct 23

Not yet reported

ARC-AGI-3

Mathematical Reasoning IQ 131 · measuredBest: #10 of 64 on MathArena4/5

BenchmarkScorePercentileRankLeader · source

MathArena

MathArena Models · Sep 22

60.6%

#10 / 64

Leader gpt-6.1-sol · 84.0%
MathArena Models · Sep 22

FrontierMath Tier 1-3

Full source leaderboard reconciliation · Oct 23

57.5%

#41 / 84

Leader gpt-6-astra · 93.7%
Full source leaderboard reconciliation · Oct 23

FrontierMath Tier 4

Full source leaderboard reconciliation · Oct 23

24.4%

#40 / 72

Leader gpt-6.1-sol · 100%
Full source leaderboard reconciliation · Oct 23

ProofBench 1.1

Full source leaderboard reconciliation v1.1 · Oct 27

56.0%

#27 / 48

Leader fable-5.1 · 100%
Full source leaderboard reconciliation v1.1 · Oct 27

Not yet reported

AIME

Academic Reasoning IQ 128 · measuredBest: #41 of 159 on CritPt4/6

BenchmarkScorePercentileRankLeader · source

CritPt

Artificial Analysis v4.3.2 · Oct 27

16.6%

#41 / 159

Leader Muse Wonder Full · 32.9%
Artificial Analysis v4.3.2 · Oct 27

GPQA Diamond

Artificial Analysis v4.3.2 · Oct 27

90.8%

#40 / 147

Leader gpt-6-astra · 96.1%
Artificial Analysis v4.3.2 · Oct 27

Humanity's Last Exam

Artificial Analysis v4.3.2 · Oct 27

38.6%

#56 / 167

Leader opus-5.5 · 61.4%
Artificial Analysis v4.3.2 · Oct 27

SciCode

Artificial Analysis v4.3.2 · Oct 27

50.3%

#63 / 167

Leader opus-5.5 · 66.9%
Artificial Analysis v4.3.2 · Oct 27

Not yet reported

MMLU-ProMMMU-Pro

Programmatic Reasoning IQ 127 · measuredBest: #39 of 104 on Terminal-Bench 4.04/6

BenchmarkScorePercentileRankLeader · source

Terminal-Bench 4.0

Artificial Analysis v4.3.2 · Oct 27

12.1%

#39 / 104

Leader Kimi K3.1 Pro · 67.2%
Artificial Analysis v4.3.2 · Oct 27

ProgramBench

Full source leaderboard reconciliation · Oct 23

8.5%

#32 / 61

Leader opus-5.5 · 65.0%
Full source leaderboard reconciliation · Oct 23

SWE-rebench

SWE-rebench · Sep 22

38.4%

#21 / 25

Leader fable-5 · 64.5%
SWE-rebench · Sep 22

IOI V2

Full source leaderboard reconciliation v2 · Oct 27

32.7%

#38 / 43

Leader Gemini 4 Argon · 100%
Full source leaderboard reconciliation v2 · Oct 27

Not yet reported

LiveCodeBenchMirrorCode

Computer Use IQ 122 · measuredBest: #28 of 70 on Arena.ai Agent Arena2/5

BenchmarkScorePercentileRankLeader · source

Arena.ai Agent Arena

1.8

#28 / 70

Leader opus-5.5 · 20.9

BrowseComp

LLM Stats BrowseComp · Sep 22

73.2%

#28 / 43

Leader Kimi K3.1 Pro · 92.1%
LLM Stats BrowseComp · Sep 22

Not yet reported

OSWorld-VerifiedMCP AtlasAgents' Last Exam

Reliability IQ 114 · measuredBest: #8 of 96 on IFBench5/6

BenchmarkScorePercentileRankLeader · source

IFBench

Artificial Analysis model leaderboard · Aug 19

79.0%

#8 / 96

Leader minimax-m3 · 82.9%
Artificial Analysis model leaderboard · Aug 19

AA Long Context Reasoning v1.1

Artificial Analysis v4.3.2 · Oct 27

79.7%

#49 / 154

Leader kimi-k3 · 88.7%
Artificial Analysis v4.3.2 · Oct 27

BullshitBench v2

Full source leaderboard reconciliation · Oct 23

39.0%

#44 / 83

Leader opus-4.8 · 94.0%
Full source leaderboard reconciliation · Oct 23

AA Omniscience

Artificial Analysis v4.3.2 · Oct 27

-14.3

#100 / 158

Leader opus-5.5 · 46.4
Artificial Analysis v4.3.2 · Oct 27

SimpleQA Verified

Full source leaderboard reconciliation · Oct 23

33.6%

#61 / 80

Leader Gemini 4 Argon · 76.0%
Full source leaderboard reconciliation · Oct 23

Not yet reported

FACTS Grounding

Knowledge Work IQ 122 · measuredBest: #38 of 124 on GDPval-AA v2.1 normalized score4/6

BenchmarkScorePercentileRankLeader · source

GDPval-AA v2.1 normalized score

Artificial Analysis v4.3.2 · Oct 20

46.7%

#38 / 124

Leader opus-5.5 · 68.3%
Artificial Analysis v4.3.2 · Oct 20

AutomationBench-AA

Artificial Analysis v4.3.2 · Oct 27

54.0%

#35 / 106

Leader Gemini 4 Argon · 77.5%
Artificial Analysis v4.3.2 · Oct 27

AA-Briefcase v1.1 Elo

Artificial Analysis v4.3.2 · Oct 27

1,257

#43 / 111

Leader sonnet-5.5 · 1,812
Artificial Analysis v4.3.2 · Oct 27

GDP.pdf

Artificial Analysis v4.3.2 · Oct 27

11.0%

#58 / 106

Leader Muse Wonder Full · 32.6%
Artificial Analysis v4.3.2 · Oct 27

Not yet reported

AA-AnalystAgentEnterpriseOps-Gym-AA

Release & availability

Facts from the publisher's announcement and documentation.

Released: Apr 24, 2026
Developer: DeepSeek
Access: Public beta through the DeepSeek API; open weights are available for local deployment
Weights: Open weightsDownloadable and self-hostable
Inputs: text
Context: 1,000,000 tokens
API price: $0.44 in · $1.32 outper 1M tokens · full pricing

What DeepSeek says

The 0731 checkpoint keeps the preview architecture and 284B/13B parameter counts, but receives new post-training focused on agent capabilities.
The API revision adds native Responses API support and Codex-oriented integration.
The canonical model ID remains deepseek-v4-flash, so dated revisions are tracked separately from the current model row.

Limitations

The July launch benchmarks use DeepSeek's own agent harness and are not substituted for independent Artificial Analysis results.
The July update applies to the V4 Flash API; DeepSeek stated that V4 Pro and its app/web models were unchanged.

Publisher statements, not AI IQ measurements.

Sourcesapi-docs.deepseek.com/updates api-docs.deepseek.com/quick_start/pricinghuggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731artificialanalysis.ai/models/deepseek-v4-flash

About this score

How much of deepseek-v4-flash's IQ is measured, and how the rest is estimated.

27/ 39benchmark results

7/ 7dimensions measured

0estimated

4/5Abstract

4/5Math

4/6Academic

4/6Coding

2/5Computer Use

5/6Reliability

4/6Knowledge Work

Bar height = share of benchmarks with resultsEstimatedNo score

Measured results are shown as published

Every score in the benchmark tables comes from its source unchanged. Composite IQ averages all seven dimensions equally.

How gaps are handled

Every dimension has at least one published result.

A missing result is not a zero

Benchmark versions and evaluation harnesses can differ between sources, so small gaps between models may not be meaningful.

Read the scoring methodology →

Compare deepseek-v4-flash

Side-by-side pages for the models it is most often weighed against.

#1 overall

gpt-6.1-sol

IQ141 20 higherEffective cost$6.30 3.7× pricier

DeepSeek #1

DeepSeek V4.1 Flash 1017

IQ128 7 higherEffective cost$2.42 1.4× pricier

Closest IQ

mimo-v2.6-flash

IQ121 same IQEffective cost$0.69 2.5× cheaper

Most similar profile

Qwen3.8 Max

IQ123 2 higherEffective cost$11.16 6.6× pricier

Similar price

grok-4.3

IQ112 9 lowerEffective cost$1.69 about the same

#2 overall

opus-5.5

IQ141 20 higherEffective cost$35.73 21.1× pricier

deepseek-v4-flashComparison model

Compare with any modelBrowse all comparisons →

Frequently asked questions

Quick answers about deepseek-v4-flash, drawn from the same data as the rest of this page.

How AI IQ scores models →Get this data from the API →

What is the IQ of deepseek-v4-flash?: deepseek-v4-flash has an estimated IQ of 121, which ranks #42 out of 176 public models on the AI IQ leaderboard. The score combines seven equally weighted reasoning dimensions using measured benchmark results and conservative estimates for missing coverage.
What is deepseek-v4-flash best at?: Among the dimensions with published benchmark results, deepseek-v4-flash ranks highest in Mathematical Reasoning (IQ 131, #30 of 176) and Abstract Reasoning (IQ 103, #46 of 201). Its lowest-ranked measured dimension is Reliability (IQ 114, #75 of 177).
What is the EQ of deepseek-v4-flash?: deepseek-v4-flash has an estimated EQ of 113. EQ measures emotional reasoning — how well a model reads and responds to emotional context — and is scored separately from IQ.
Which benchmarks has deepseek-v4-flash been tested on?: deepseek-v4-flash has published results on 27 of the 39 benchmarks behind AI IQ's Composite IQ, including IFBench, MathArena, CritPt, GPQA Diamond and GDPval-AA v2.1 normalized score.
How much does deepseek-v4-flash cost?: deepseek-v4-flash costs $0.44 per million input tokens and $1.32 per million output tokens. Adjusted for how many tokens it actually consumes on real tasks, its effective cost works out to $1.69 per million input-plus-output tokens.
Is deepseek-v4-flash open source?: Yes — deepseek-v4-flash is an open-weights model: its trained weights are published, so it can be downloaded and self-hosted.
Who developed deepseek-v4-flash?: deepseek-v4-flash was developed by DeepSeek, based in China.
When was deepseek-v4-flash released?: deepseek-v4-flash was publicly released on April 24, 2026.
How large is the context window of deepseek-v4-flash?: deepseek-v4-flash supports a context window of 1,000,000 tokens — the amount of text it can consider in a single request.
How fast is deepseek-v4-flash?: In independent measurements, deepseek-v4-flash generates about 142 tokens per second, with a typical response time of about 19 seconds.