DeepSeek ships V4.1 Flash, the smallest model in its new multimodal family

DeepSeek has released DeepSeek-V4.1-Flash, the smallest model built on the company’s new architecture and its first in that family with native multimodal visual understanding, meaning it can take images as input alongside text.

The model arrived on 10 September and is available through the DeepSeek API, according to the company’s changelog.

What’s new

DeepSeek says the new architecture is designed for:

• a higher capability ceiling as models scale up,
• faster inference and higher throughput, and
• scaling to larger models in the same family.

“Flash” models are the fast, inexpensive tier: the kind developers reach for when they need to process large volumes of requests, such as classifying documents or reading screenshots, where cost per call matters more than peak intelligence.

V4 Pro stays

In a separate update, DeepSeek said that “in response to user demand” it will continue offering API access to DeepSeek V4 Pro after 14 September, with billing unchanged. Keeping older models available matters for businesses that have tuned their products around a specific model’s behaviour and cannot switch overnight.

The price pressure continues

DeepSeek has built its reputation on capable models at low prices. In August, Axios described a previous DeepSeek release as accelerating AI’s “race to zero” on pricing. A cheap, fast model that can also read images puts more pressure on rivals’ small-model tiers, and gives developers one more option when choosing how to feed their own data to a model.