DeepSeek V4 Flash

by DeepSeek

DeepSeek V4 Flash 0731 is the smaller, more efficient member of the DeepSeek V4 family. It has 284B total parameters, with 13B active per token, and a 1,048,576-token context window. Hybrid attention helps keep long prompts cheaper to process. You can also adjust reasoning effort to trade speed and token use for more deliberate answers.

Best for: Low-cost text agents with long prompts and reusable prefixes.

1M context

Text
Reasoning
Tool use

Model weights

Try it in the playground

Pricing

USD per 1M tokens. The 1h and 24h batch windows will be launching in the coming weeks.

Window	Input	Cached input	Output
asap	$0.04	$0.02	$0.21
<1h-25% off	$0.03	$0.02	$0.16
<24h-50% off	$0.02	$0.01	$0.11

What one request costs

Prompt94Ktokens

Cached prefix66Ktokens

Reply2Ktokens

$0.0028

per request to DeepSeek V4 Flash

Prompt: 40.9%
Cached: 47.7%
Reply: 11.4%

The same request, across other models