DeepSeek unveiled an experimental AI model that can handle visual prompts, Bloomberg Technology reports. The Hangzhou-based company said the tool nears the performance of an advanced model from Anthropic PBC, with Bloomberg’s headline identifying that rival model as Anthropic’s Opus 4.8. According to Bloomberg, the release is an experimental version of DeepSeek’s flagship text-only V4 Flash model. The test model adds multimodal capabilities, meaning it can analyze and act on visual inputs such as images and screenshots, not only text prompts. That is the concrete change: V4 Flash is being tested beyond its text-only base into vision-enabled use cases. Bloomberg’s report does not include benchmark scores, testing methodology, release terms, pricing, or availability details, so the performance comparison should be read as DeepSeek’s claim rather than an independently verified result. The named comparison matters because Anthropic is the US rival cited in Bloomberg’s report, and Opus 4.8 is presented as the advanced model DeepSeek is trying to approach. For now, the defensible read is narrow: DeepSeek has put a test multimodal model into the field and is positioning it against a high-end Anthropic system. Who benefits: DeepSeek benefits if the experimental model proves competitive in multimodal tasks. Users evaluating visual-prompt workflows may get another option to test, pending availability details. Who's exposed: Anthropic is the named benchmark in Bloomberg’s report. Broader competitive implications are too early to assess without independent performance data.