DeepSeek Launches V4-Flash-Vision-Exp Multimodal Model with Token-Based API Pricing
DeepSeek has launched its experimental multimodal vision model DeepSeek-V4-Flash-Vision-Exp on its API platform. The model delivers a substantial improvement over V4 Flash on vision-based Agent benchmarks, with multimodal Agent capability approaching Opus-4.8. Images are billed per token, with each image consuming up to 384 tokens at the same rate as V4-Flash. An official demonstration shows the model generating a commercial Tibet self-drive tour presentation with clean, refined visuals.
On August 21, DeepSeek announced the launch of its multimodal vision understanding model, DeepSeek-V4-Flash-Vision-Exp, on its API platform. The experimental model offers multimodal adaptation capabilities and supports users in completing practical work scenarios through a range of Agent tools.
On Agent benchmarks requiring visual understanding, the new model posts a significant improvement over V4 Flash, with multimodal Agent capability approaching that of Opus-4.8. An official use case shows the model generating a commercially tailored presentation for a Tibet self-drive tour, featuring clean and refined visuals.
In terms of pricing, images uploaded through the API are converted into token-based billing, with each image consuming up to 384 tokens. The billing rate is identical to that of the V4-Flash model.
Why this event matters
The event has a measured impact on 1 industry. The strongest current signal is positive for Artificial Intelligence, with intensity 60/100 and 70% confidence over a short term horizon.
Impact figures are analytical estimates that combine direction, intensity, confidence and event importance. They are not investment advice.