Chinese AI startup DeepSeek has unveiled an experimental multimodal model designed to rival U.S. competitor Anthropic, expanding its capabilities beyond text to include visual inputs such as images and screenshots.
The new model, named DeepSeek-V4-Flash-Vision-Exp, builds on the company’s flagship text-only V4 Flash model, which launched earlier this year. It retains parity with V4 Flash in text-based tasks, including reasoning and world knowledge, while introducing multimodal agent functionality that allows it to process and act on visual prompts without constant human oversight.
On multimodal agent benchmarks, DeepSeek-V4-Flash-Vision-Exp narrows the performance gap with Anthropic’s Opus-4.8, a leading advanced model in the segment. The Chinese firm claims the new model delivers a significant leap in multimodal capabilities compared with its predecessor.
The model is now accessible via the DeepSeek API Platform, with images tokenized for billing purposes at up to 384 tokens each at V4 Flash pricing. Input options include base64-encoded images, external URLs, or the Files API, enabling flexible integration for developers.
The announcement was made on Friday, August 21, 2026, as part of DeepSeek’s ongoing efforts to expand its AI offerings amid intensifying competition in the global artificial intelligence market.
