Advertise here
#53

DeepSeek V4.1 Flash

DeepSeek Release: 2026-09-10 Tested on: 2026-09-10 12:40 deepseek/deepseek-v4.1-flash::medium
748B total (16B active)MoEOpen source
(max) (high) (medium) (low) (none)

Summary

DeepSeek V4.1 Flash scores 8.5 on AI BENCHY and ranks #53. It has 9.5 reliability, a 72.7% pass rate, $0.272 total cost, and 32.22s average response time.

What makes DeepSeek V4.1 Flash unique: Its total benchmark cost is unusually low for its score range.

Model facts

Researched on 2026-09-10

Reported
Parameters
748B total (16B active)
Architecture
MoE
Availability
Open source
License
MIT

Total is 552B backbone plus 196B Engram conditional-memory parameters. Active count is 16B during decode; prefill activates 8B per token. The exact checkpoint includes MIT-licensed weights and a runnable inference implementation.

Parameter source Availability source

Score

8.5

Consistency

9.3

Reliability

9.5

Total Cost (Current Price)

$0.272

Total Output Tokens

430,816

Total Input Tokens

88,508

Input Price

$0.150 / 1M

Output Price

$0.600 / 1M

Tests Correct

Wrong Tests: 7

Attempt pass rate: 72.7%

Flaky tests

2

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

Response Time (avg)

32.22s

Response Time (max): 276.31s

Response Time (total): 708.76s

Wrong answer: 5 API error: 1 Did not follow instructions: 1

Hamster playing table tennis

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

#53 DeepSeek V4.1 Flash

medium
Cost
$0.014
Time
34.7s
Tokens
11,541 tok

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 V4.1 FlashmediumvsDeepSeek V4.1 FlashhighDeepSeek V4.1 FlashmediumvsGrok 4.5mediumDeepSeek V4.1 FlashmediumvsDeepSeek V4.1 FlashlowDeepSeek V4.1 FlashmediumvsMuse Spark 1.2lowDeepSeek V4.1 FlashmediumvsGPT-5.4mediumDeepSeek V4.1 FlashmediumvsGrok 4.6mediumDeepSeek V4.1 FlashmediumvsQwen3.8 MaxlowDeepSeek V4.1 FlashmediumvsGrok 4.5lowDeepSeek V4.1 FlashmediumvsQwen3.6 Max PreviewmediumDeepSeek V4.1 FlashmediumvsGPT-5.2medium

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 6.5 10.0
Domain specific 2.9 7.2
General Intelligence 10.0 10.0
Instructions following 10.0 10.0
Puzzle Solving 8.7 7.7
Tool Calling 10.0 10.0
Trivia 3.0 10.0

Compared models