AI BENCHY
Advertise here
#35

DeepSeek V4 Flash 0423

DeepSeek Release: 2026-04-24 Tested on: 2026-04-24 09:19 deepseek/deepseek-v4-flash::high
284B total (13B active)MoEOpen source
(high) (none)

Summary

DeepSeek V4 Flash 0423 scores 7.8 on AI BENCHY and ranks #35. It has N/A reliability, a 79.6% pass rate, $0.019 total cost, and 39.75s average response time.

What makes DeepSeek V4 Flash 0423 unique: It stands out most in Data parsing and extraction, where it ranks #1, while Puzzle Solving is its weakest area at #18. Its total benchmark cost is unusually low for its score range.

Model facts

Researched on 2026-08-12

Reported
Parameters
284B total (13B active)
Architecture
MoE
Availability
Open source
License
MIT
Parameter source Availability source

Score

7.8

Consistency

7.8

Reliability

N/A

Total Cost (Current Price)

$0.019

Total Output Tokens

48,916

Total Input Tokens

0

Input Price

$0.140 / 1M

Output Price

$0.280 / 1M

Tests Correct

Wrong Tests: 7

Attempt pass rate: 79.6%

Flaky tests

5

Flaky tests had mixed outcomes across runs (at least one pass and one fail).

Response Time (avg)

39.75s

Response Time (max): 116.28s

Response Time (total): 715.44s

Did not follow instructions: 3 Wrong answer: 3 Timed out: 1

Hamster playing table tennis

Prompt: Create a detailed SVG illustration of a hamster playing table tennis.

#35 DeepSeek V4 Flash

high
Cost
$0.003
Time
93.1s
Tokens
7,926 tok

Run history

Tested on Score Reliability Tests Correct Total Cost Compare
2026-08-14 03:03 Re-test 7.6 10.0 $0.066 ↑ Compare
2026-07-16 23:17 New test added 7.7 10.0 $0.042 ↓ Compare
2026-06-04 14:24 Re-test 7.7 10.0 $0.029 ↓ Compare
2026-06-04 13:02 New test added 7.7 10.0 $0.028 ↓ Compare
2026-05-22 00:39 Suite changed 7.4 10.0 $0.038 Compare
2026-04-25 21:41 Re-test 7.8 N/A $0.031 Compare
2026-04-24 09:19 Initial run 7.8 N/A $0.019 Current run

Run comparison

RunBenchmark coverageScoreConsistencyReliabilityTests CorrectFlaky testsTotal Output TokensTotal Input TokensTotal CostResponse Time (avg)
2026-04-24 09:19 · Initial run54/54 attempts7.87.8N/A11/18548,9160$0.01939.75s
2026-05-22 00:39 · Suite changed60/60 attempts7.48.010.011/205126,8690$0.03846.28s
Difference—+0.4-0.200-779530-$0.019-6531ms

Benchmark coverage differs: 54/54 attempts (Target: 3 repeats per test) versus 60/60 attempts (Target: 3 repeats per test). Totals and repeat-sensitive metrics are not directly comparable.

These two runs used different benchmark suites, so the deltas reflect both model changes and suite changes.

Charts

Choose the first model, then click a second model to open a side-by-side page.

Total Output Tokens

Score vs Total Output Tokens

Quick Compare

DeepSeek V4 Flash 0423highvsDeepSeek V4 ProhighDeepSeek V4 Flash 0423highvsGPT-5.6 LunahighDeepSeek V4 Flash 0423highvsClaude Opus 4.6mediumDeepSeek V4 Flash 0423highvsSolar Pro 4xhighDeepSeek V4 Flash 0423highvsSeed-2.0-CodelowDeepSeek V4 Flash 0423highvsNemotron 3 UltramediumFree AvailableDeepSeek V4 Flash 0423highvsGLM 5mediumDeepSeek V4 Flash 0423highvsClaude Opus 5noneDeepSeek V4 Flash 0423highvsQwen3.7 FlashhighDeepSeek V4 Flash 0423highvsGPT-5.4 Nanomedium

Category Breakdown

Category Score Consistency Tests Correct
Anti-AI Tricks 8.3 10.0
Coding 10.0 10.0
Combined 10.0 10.0
Data parsing and extraction 10.0 10.0
Domain specific 4.1 4.4
General Intelligence 6.1 3.1
Instructions following 10.0 10.0
Puzzle Solving 6.4 4.5
Tool Calling 10.0 10.0

Compared models