DeepSeek Unveils Test Model to Rival Anthropic’s Opus 4.8

DeepSeek unveiled an experimental AI model that can understand visual prompts, saying the tool nears the performance of an advanced model by US rival Anthropic PBC.

The new release is an experimental version of DeepSeek’s flagship text-only V4 Flash model that adds multimodal capabilities, meaning it can also analyze and act on visual prompts like images and screenshots, the Hangzhou-based company said on Friday.

“This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, reasoning, and world knowledge,” the company wrote on X.

The model’s performance is “close to” Anthropic PBC’s advanced Opus 4.8 model on tests of its multimodal agentic capabilities, where a model can take action without constant oversight or prompting, the company said.

Read More: US Lead in the AI Race With China Is Rapidly Narrowing

DeepSeek has released multimodal and visual models before, such as its DeepSeek-VL family. But this new offering, called DeepSeek-V4-Flash-Vision-Exp, adds the ability to interpret images and carry out tasks to its most cutting-edge model series. Users can access the model through DeepSeek’s API.

Chinese firms are locked in a heated rivalry with US developers at the leading edge of AI, with their models often offering similar performance for a fraction of the price. The arrival of DeepSeek’s flagship model earlier this year reset expectations for what low-cost, open-weight models could achieve.

Read More: China’s AI Blitz Creates ‘Death Zone’ for Rival US Model Makers