Key facts
- DeepSeek has launched an experimental multimodal AI model, DeepSeek-V4-Flash-Vision-Exp.
- The new model can process visual inputs and retain text capabilities like reasoning and general knowledge.
- DeepSeek claims the model's multimodal agentic performance is comparable to Anthropic's Opus 4.8.
- The model is offered through an API for developers.
- Anthropic is reportedly planning an IPO, with preparations being made by investment banks.
China's DeepSeek has unveiled an experimental multimodal AI model, DeepSeek-V4-Flash-Vision-Exp, designed to process visual inputs alongside text. This new model is an extension of the company's V4 Flash platform and aims to compete with leading AI systems, including Anthropic's Opus 4.8, in multimodal agentic performance.
According to DeepSeek, the experimental model retains the text-based capabilities of its V4 Flash predecessor, such as reasoning, agentic tasks, and general knowledge. The company stated that in evaluations of multimodal agentic performance, its new system is similar to Anthropic's Opus 4.8, which assesses AI systems performing tasks with minimal human involvement using visual information.
This development occurs amid intensifying competition among major AI developers. DeepSeek, a prominent AI startup in China, has been expanding its AI model portfolio. The launch also follows reports that Anthropic is preparing for a potential initial public offering (IPO), which could be a significant market event, potentially rivaling the scale of SpaceX's previous offering.
Investment banks including Morgan Stanley, Goldman Sachs, and JPMorgan are reportedly making preparations for Anthropic's potential IPO, with some reports suggesting it could file as early as this month. DeepSeek's existing DeepSeek-VL series already specializes in visual-language aspects, and the new model integrates similar features into its newer V4 model family.