DeepSeek-V4-Flash Benchmark Details

DeepSeek-V4-Flash currently shows benchmark results led by LiveCodeBench (6 / 250, score 91.60), IF Bench (11 / 282, score 79.20), HLE (43 / 563, score 51.50). This page also compares it with 3 competitor models and 2 predecessor or same-series models, including performance and pricing views when available.

Benchmark Results

DeepSeek-V4-Flash

Benchmark Results

General Knowledge

16 evaluations

Benchmark / mode

Score

Rank/total

MMLU Pro

Standard Mode

47 / 176

MMLU Pro

High

86.40

18 / 176

MMLU Pro

Max

86.20

19 / 176

LiveBench

Standard Mode

65.48

52 / 117

HLE

Standard Mode

8.10

418 / 563

HLE

Standard Mode

7.80

424 / 563

HLE

High

30.30

200 / 563

HLE

High

29.40

210 / 563

HLE

HighTools

40.30

130 / 563

HLE

Max

34.80

171 / 563

HLE

Max

34.80

171 / 563

HLE

MaxTools

51.50

43 / 563

HLE

Thinking Level · Extra HighTools

45.10

82 / 563

CritPt

Standard Mode

0.30

182 / 200

CritPt

High

3.40

109 / 200

CritPt

Max

7.10

87 / 200

Other

3 evaluations

Benchmark / mode

Score

Rank/total

GPQA Diamond

Standard Mode

71.20

311 / 462

GPQA Diamond

High

86.70

136 / 462

GPQA Diamond

Max

89.40

96 / 462

Coding and Software Engineer

21 evaluations

Benchmark / mode

Score

Rank/total

CodeForces

High

2816

6 / 21

CodeForces

Max

3289

3 / 21

Text Arena (Coding)

High

1576.54

6 / 35

LiveCodeBench

Standard Mode

55.20

151 / 250

LiveCodeBench

High

88.40

16 / 250

LiveCodeBench

Max

91.60

6 / 250

SWE-bench Verified

Standard ModeTools

73.70

45 / 116

SWE-bench Verified

HighTools

78.60

23 / 116

SWE-bench Verified

Thinking Level · Extra HighTools

20 / 116

SWE-bench Multilingual

Standard ModeTools

69.70

24 / 30

SWE-bench Multilingual

HighTools

70.20

22 / 30

SWE-bench Multilingual

Thinking Level · Extra HighTools

73.30

17 / 30

NL2Repo-Bench

MaxTools

54.20

12 / 16

DeepSWE

MaxTools

53.32

56 / 85

SWE-Bench Pro - Public

Standard ModeTools

49.10

50 / 62

SWE-Bench Pro - Public

HighTools

52.30

42 / 62

SWE-Bench Pro - Public

Thinking Level · Extra HighTools

52.60

40 / 62

SciCode

High

40.20

101 / 130

SciCode

Max

45.30

89 / 130

WeirdML v2

HighTools

43.76

44 / 52

SEC-Bench Pro

MaxTools

30.90

6 / 6

Writing and Creative Capabilities

1 evaluations

Benchmark / mode

Score

Rank/total

Creative Writing

Standard Mode

1555.70

46 / 106

Common Sense Reasoning

2 evaluations

Benchmark / mode

Score

Rank/total

SimpleBench

Standard Mode

61.10

29 / 93

SimpleBench

Standard Mode

46.30

57 / 93

Agent Level Benchmark

9 evaluations

Benchmark / mode

Score

Rank/total

τ²-Bench - Telecom

Standard ModeTools

94.40

35 / 264

τ²-Bench - Telecom

HighTools

95.60

26 / 264

τ²-Bench - Telecom

MaxTools

32 / 264

Terminal Bench Hard

Standard ModeTools

34.10

82 / 244

Terminal Bench Hard

HighTools

38.60

57 / 244

Terminal Bench Hard

MaxTools

35.60

70 / 244

τ³-Banking

HighTools

26.20

71 / 164

τ³-Banking

MaxTools

30.90

57 / 164

Agents' Last Exam

MaxTools

25.20

16 / 19

Instruction Following

3 evaluations

Benchmark / mode

Score

Rank/total

IF Bench

Standard Mode

47.20

170 / 282

IF Bench

High

73.50

43 / 282

IF Bench

Max

79.20

11 / 282

AI Agent - Information Search

2 evaluations

Benchmark / mode

Score

Rank/total

BrowseComp

HighTools

53.50

42 / 57

BrowseComp

Thinking Level · Extra HighTools

73.20

30 / 57

AI Agent - Tool Usage

10 evaluations

Benchmark / mode

Score

Rank/total

CyberGym

MaxTools

76.70

7 / 8

Toolathlon-Verified

MaxTools

70.30

11 / 11

Terminal-Bench 2.1

HighTools

56.90

117 / 192

Terminal-Bench 2.1

MaxTools

61.80

105 / 192

Terminal Bench 2.0

Standard ModeTools

49.10

36 / 48

Terminal Bench 2.0

HighTools

56.60

28 / 48

Terminal Bench 2.0

Thinking Level · Extra HighTools

56.90

26 / 48

Terminal-Bench 3.0

MaxTools

7.60

11 / 11

Terminal-Bench 4.0

HighTools

60 / 86

Terminal-Bench 4.0

MaxTools

2.50

63 / 86

Text Embedding

2 evaluations

Benchmark / mode

Score

Rank/total

Context Arena

Standard Mode

26.47

117 / 126

Context Arena

Thinking Enabled

69.42

66 / 126

Math and Reasoning

7 evaluations

Benchmark / mode

Score

Rank/total

AIME 2026

MaxTools

95.83

10 / 29

HMMT Feb 2026

MaxTools

93.94

6 / 10

IMO-AnswerBench

Standard Mode

41.90

23 / 24

IMO-AnswerBench

High

85.10

12 / 24

IMO-AnswerBench

Max

88.40

7 / 24

MathArena Apex

Max

58.60

7 / 17

MathArena Apex

MaxTools

27.08

12 / 17

Long Context

3 evaluations

Benchmark / mode

Score

Rank/total

AA-LCR

Standard Mode

41.70

156 / 170

AA-LCR

High

95 / 170

AA-LCR

Max

74.30

86 / 170

Productivity Knowledge

7 evaluations

Benchmark / mode

Score

Rank/total

GDPval-AA v2

HighTools

1083

84 / 105

GDPval-AA v2

MaxTools

1116

79 / 105

AA-Briefcase

HighTools

1052

57 / 83

AA-Briefcase

MaxTools

833

75 / 83

Harvey Lab-AA

MaxTools

81.33

31 / 43

AutomationBench

MaxTools

37.70

10 / 17

AA-AnalystAgent

MaxToolsInternet

16 / 29

Multimodal Understanding

2 evaluations

Benchmark / mode

Score

Rank/total

GDP.pdf

High

10.40

78 / 118

GDP.pdf

Max

10.80

77 / 118

Other

1 evaluations

Benchmark / mode

Score

Rank/total

ExploitGym (budget unspecified)

MaxTools

1.80

6 / 6

Claw-style Agent Evaluation

1 evaluations

Benchmark / mode

Score

Rank/total

PinchBench v2

unknown

81.74

8 / 45

Compare with other models

Competitor Comparison

Benchmark scores for DeepSeek-V4-Flash compared against top models in its class

DeepSeek-V4-FlashGLM 5.1 Qwen3.6-27B Gemini 3.5 Flash

Benchmark categories:

The chart shows each model’s highest score per benchmark within the current filter. Out-of-100 benchmarks use raw heights; out-of-range benchmarks are scaled within that benchmark while labels keep the original scores.

12 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.

Benchmark	DeepSeek-V4-FlashCurrent	GLM 5.1	Qwen3.6-27B	Gemini 3.5 Flash
CritPt Score 综合评估	7.10Thinking Level · High	4.60Thinking Enabled	1.10Thinking Enabled	13.10Thinking Level · High
HLE Accuracy 综合评估	51.50Thinking Level · High ｜ Tools	52.30Thinking Enabled ｜ Tools	24.00Thinking Enabled	42.70Thinking Level · High
LiveBench Accuracy 综合评估	65.48Standard Mode	70.18Standard Mode	64.03Standard Mode	74.64Thinking Level · High
MMLU Pro Accuracy 综合评估	86.40Thinking Level · High	--	86.20Thinking Enabled	--
GPQA Diamond Accuracy 科学与综合推理	89.40Thinking Level · High	86.20Thinking Enabled	87.80Thinking Enabled	92.80Thinking Level · High
DeepSWE Pass@1 (DeepSWE v1.1) 编程与软件工程	53.32Thinking Level · High ｜ Tools	--	--	37.00Thinking Level · Medium ｜ Tools
LiveCodeBench Pass @K 编程与软件工程	91.60Thinking Level · High	--	83.90Thinking Enabled	--
SciCode Score 编程与软件工程	45.30Thinking Level · High	44.80Thinking Enabled	42.80Thinking Enabled	53.90Thinking Level · High
SWE-bench Multilingual Accuracy 编程与软件工程	73.30Thinking Level · Extra High ｜ Tools	--	71.30Thinking Enabled ｜ Tools	--
SWE-Bench Pro - Public Accuracy 编程与软件工程	52.60Thinking Level · Extra High ｜ Tools	58.40Thinking Enabled ｜ Tools	53.50Thinking Enabled ｜ Tools	55.10Thinking Level · High ｜ Tools
SWE-bench Verified Accuracy 编程与软件工程	79.00Thinking Level · Extra High ｜ Tools	--	77.20Thinking Enabled ｜ Tools	--
Text Arena (Coding) Arena Score 编程与软件工程	1576.54Thinking Level · High	1534.00Standard Mode	--	--

21 additional benchmarks remain in the chart above.

Standard API Pricing: DeepSeek-V4-Flash vs. Peer Models

Shows standard text input and output pricing side by side for each model. If extended-context pricing exists, the chart keeps the base rate and explains the threshold below.

Source: DataLearnerAI. Standard text prices shown here use the default supplier. · USD / 1M tokens

Model	Supplier	Standard input	Standard output	Base price applies to
DeepSeek-V4-Flash	DeepSeek-AI	$0.14 / 1M tokens	$0.28 / 1M tokens	—
GLM 5.1	智谱AI	$1.4 / 1M tokens	$4.4 / 1M tokens	—
Gemini 3.5 Flash	DeepMind	$1.5 / 1M tokens	$9 / 1M tokens	—

Version History

How each version of the DeepSeek-V4-Flash series stacks up on benchmark tests

DeepSeek-V4-FlashDeepSeek V4 DeepSeek V3.2

Benchmark categories:

12 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.· Click a row to view its trend chart.

Benchmark	DeepSeek-V4-FlashCurrent	DeepSeek V3.2
CritPt Score 综合评估	7.10Thinking Level · High	2.90Thinking Enabled
HLE Accuracy 综合评估	51.50Thinking Level · High ｜ Tools	25.10Thinking Enabled
LiveBench Accuracy 综合评估	65.48Standard Mode	62.20Thinking Enabled
GPQA Diamond Accuracy 科学与综合推理	89.40Thinking Level · High	82.40Thinking Enabled
CodeForces Accuracy 编程与软件工程	3289.00Thinking Level · High	2386.00Thinking Enabled
LiveCodeBench Pass @K 编程与软件工程	91.60Thinking Level · High	83.30Thinking Enabled
SWE-Bench Pro - Public Accuracy 编程与软件工程	52.60Thinking Level · Extra High ｜ Tools	40.90Thinking Enabled
SWE-bench Verified Accuracy 编程与软件工程	79.00Thinking Level · Extra High ｜ Tools	73.10Thinking Enabled ｜ Tools
Creative Writing Elo、大模型评判两两对战写作和创作	1555.70Standard Mode	1511.20Standard Mode
Terminal Bench Hard Accuracy Agent能力评测	38.60Thinking Level · High ｜ Tools	35.60Thinking Enabled ｜ Tools
τ²-Bench - Telecom Accuracy Agent能力评测	95.60Thinking Level · High ｜ Tools	90.60Thinking Enabled ｜ Tools
IF Bench Accuracy 指令跟随	79.20Thinking Level · High	60.70Thinking Enabled

5 additional benchmarks remain in the chart above.

Single-Benchmark Version Trend

Viewing: CritPt · 综合评估

Benchmark

NormalNormal + ToolsThinkingThinking + ToolsDeepDeep + Tools

X-axis shows model and release date, Y-axis shows score; solid lines connect the same mode across versions, while dotted guides align modes within the same generation.

Standard API Pricing Across the DeepSeek-V4-Flash Series

Shows standard text input and output pricing side by side for each model. If extended-context pricing exists, the chart keeps the base rate and explains the threshold below.

Source: DataLearnerAI. Standard text prices shown here use the default supplier. · USD / 1M tokens

Model	Supplier	Standard input	Standard output	Base price applies to
DeepSeek-V4-Flash	DeepSeek-AI	$0.14 / 1M tokens	$0.28 / 1M tokens	—
DeepSeek V3.2	DeepSeek-AI	$0.28 / 1M tokens	$0.42 / 1M tokens	—