DeepSeek Launches V4.1-Flash and Plans to Replace V4-Pro API Traffic

DeepSeek has released V4.1-Flash, a new AI model with native visual understanding, and says it will soon direct requests from its V4-Pro API endpoint to the new model. The move puts a smaller member of DeepSeek’s new architecture family in line to become the default for customers who had selected the company’s higher-tier option.

The release is available through the DeepSeek API as deepseek-flash. DeepSeek calls V4.1-Flash the smallest model in its new architecture family and says the design is intended to raise the model’s capability ceiling while improving inference speed, throughput and the ability to scale to larger models. Native multimodal support means the model can accept visual as well as text input.

One new model, two different promises

For prospective users, DeepSeek’s first promise is a conventional product claim: a faster, more capable and higher-throughput Flash model. The second is more consequential. The company says extensive testing found V4.1-Flash ahead of V4-Pro on performance, cost, speed and total time. That is DeepSeek’s assessment of its own tested workloads, rather than an independent ranking of the two systems.

A migration, not just another endpoint

The switchover changes the practical meaning of the launch for existing API customers. Rather than asking every V4-Pro user to choose a successor, DeepSeek plans to make V4.1-Flash the destination automatically until it releases V4.1-Pro. The planned switch trades continued access to a named higher-tier endpoint for DeepSeek’s claim that its Flash replacement is the stronger deal.

What API users need to know

• New calls can use deepseek-flash to reach V4.1-Flash.
• The older deepseek-v4-flash and deepseek-v4-flash-vision-exp names temporarily point to V4.1-Flash for compatibility.
• DeepSeek has retired the prior V4-Flash and V4-Flash-Vision-Exp models.

Flash moves closer to the flagship

That positioning sharpens a contrast already present in the V4 family. When DeepSeek introduced V4 in April, it described V4-Flash as the faster, more economical alternative and V4-Pro as the model aimed at the highest performance. Both models were presented with a 1-million-token context window. V4.1-Flash now carries the Flash label, but DeepSeek says it can surpass V4-Pro across the measures that matter for its planned replacement.

DeepSeek has also reduced API prices with the release, though its public changelog directs users to a separate pricing page for the details. Its headline comparisons of V4.1-Flash with V4-Pro should therefore be read as company-supplied performance and cost claims. The announcement establishes DeepSeek’s product direction and migration plan; it does not by itself establish how the two models will compare on every customer workload.

The launch arrives as DeepSeek prepares for a potential initial public offering on Shanghai’s technology-focused STAR Market, according to Reuters. For now, the immediate test is less about that prospective listing than the forced comparison embedded in the API: whether customers who chose V4-Pro find the new Flash default suitable once the routing change begins.