Skip to content
trckr.ai

Loading…

Learn · Compare

Compare two models.

Side by side

Swap

DeepSeek V4 Pro 0813

DeepSeek's V4 Pro 0813 is a very large official release, free to download under the MIT license, aimed at developers building tools that carry out multistep tasks.

Release page →

DeepSeek V4 Flash 0731

DeepSeek's updated free-to-download model that is better at completing agent-style tasks and coding, free for anyone to use under the MIT license.

Release page →

The basics

Released

Aug 13
Jul 31

What it does

Text and chat: Writes, answers questions and holds a conversation.Thinks first: Can reason step by step before it answers: slower, but better at maths, code and tricky questions.
Text and chat: Writes, answers questions and holds a conversation.Thinks first: Can reason step by step before it answers: slower, but better at maths, code and tricky questions.Sees images: Understands pictures, screenshots and charts you share with it.

Knowledge cutoff

The latest events it learned about

Not published
May 2025

Download

FreeOpen weights
FreeOpen weights

Where to use it

DeepSeek's app: DeepSeek ↗
DeepSeek's app: DeepSeek ↗

Price and limits, for developers

Input

Per million tokens

$0.66
$0.15Cheaper

Output

Per million tokens

$1.98
$0.60Cheaper

Cached input

Per million tokens

$0.022
$0.003Cheaper

Context window

How much it takes in at once

1.05M tokens about 790,000 words
1.05M tokens about 790,000 words

Longest answer

393K tokens about 290,000 words
393K tokens about 290,000 words

How they score

Combined test score

Epoch Capabilities Index · Epoch AI

156#38 of 237
154#49 of 237

Too close to call: DeepSeek V4 Pro 0813 and DeepSeek V4 Flash 0731 are within the margin of error.

Chat answers, by people's votes

LMArena text, adjusted for style · LMArena

1,464#50 of 86Higher
1,439#85 of 86

In blind votes, people pick DeepSeek V4 Pro 0813 over DeepSeek V4 Flash 0731 about 54% of the time.

Building apps, by people's votes

LMArena WebDev · LMArena

1,581#37 of 85
1,582#36 of 85

Neck and neck: in blind votes, people pick each about as often as the other.

Green marks the better of the two on each row, red the other. Rows where one side isn't listed, or where the gap is too close to call, stay plain.

Popular comparisons

Not sure where to start? Which AI should I use? gives a pick for what you want to do. How to read AI rankings explains the numbers.

Rankings from LMArena and Epoch AI, both CC BY 4.0, updated . Developer prices, context windows and knowledge cutoffs from the labs' own listings on models.dev and from OpenRouter. The app links go to each maker's own app when no source says exactly where the model is available.

Free alerts

Know the moment a new model drops.

Pick one or several. Each has its own settings for what you hear about.