DeepSeek V4.1 Flash

by DeepSeek ~ Fast, low-cost agentic coding and visual analysis

DeepSeek V4.1 Flash is a 552B-parameter mixture-of-experts model with a causal encoder-decoder architecture: 8B parameters activate for input tokens and 16B for output tokens. It combines fast, low-cost agentic coding and reasoning with native visual understanding, tool use, and a 1M-token context window. Its substantially smaller KV cache also lowers the cost of long-running, cache-heavy workloads.

1M context

Text
Reasoning
Tool use
Image input
Vision

Model weights

Try it in the playground

Pricing

USD per 1M tokens.

Window	Input	Cached input	Output
asap	$0.20	$0.01	$1.00

What one request costs

Prompt90Ktokens

Cached prefix63Ktokens

Reply2Ktokens

$0.0075

per request to DeepSeek V4.1 Flash

Prompt: 71.7%
Cached: 8.4%
Reply: 19.9%

The same request, across the lineup