A cheap option for developers running coding agents or long documents, with peak-hour prices doubling and open weights for self-hosting.

Kenning's take

What it does

DeepSeek V4.1 Flash is a 552 billion parameter mixture-of-experts model that activates only 8 billion parameters for input and 16 billion for output. It understands images natively, handles up to one million tokens, and uses a much smaller memory cache than V4 Flash, which cuts costs for agent work. It is the default Flash model in the DeepSeek API and the weights are on Hugging Face under the MIT licence.

• 552B-parameter MoE with 8B active for input and 16B for output
• Native image understanding
• One million token context
• KV cache about a quarter the size of V4 Flash
• Adjustable reasoning effort
• Open weights under MIT licence

Who it suits. Developers who want a cheap, capable model for coding agents and long-document work through the API or self-hosting.

Key facts

Category

AI models

Part of

DeepSeek

Made by

DeepSeek

Released

10 September 2026

How to get it

in the DeepSeek API as deepseek-flash, and as open weights on Hugging Face

Works on

API, open weights

Free plan

open weights, MIT licence

API price

off-peak $0.15 per million input tokens (cache miss), $0.003 cache hit, $0.60 per million output; peak $0.30 input, $0.006 cache hit, $1.20 output

FAQ

Facts from the maker's own pages and the sources listed. Checked 2026-10-11.