DeepSeek Unleashes New AI Model

DeepSeek released an experimental multimodal model. The company claims that it brings its performance close to Anthropic’s Opus 4.8 model on tests of its agentic capabilities.

The new model, named DeepSeek-V4-Flash-Vision-Exp, is available through the Chinese AI developer’s API, adding image interpretation abilities to its V4-Flash model. The system can process images alongside text, allowing it to interpret screenshots, charts and other visual information. DeepSeek said that the model matches V4-Flash on text capabilities, and makes a major leap over V4-Flash on multimodal agent benchmarks. The company added that its model’s multimodal agent performance is close to Opus-4.8.

It supports JPEG, PNG, GIF, and WebP images and can process up to 600 images in a single request, according to DeepSeek. The company also released an update to its DeepSeek Harness tool with support for the new model. DeepSeek has previously developed multimodal models, including its DeepSeek-VL family.

Chinese AI developers are notably increasingly taking the fight to US rivals on model performance and accessibility. Earlier this month, ByteDance was reportedly developing a model with as many as 10 trillion parameters, which could rival Anthropic’s powerful Mythos AI model. Meanwhile, last month China warned that it would take all necessary measures to protect its interests after the US threatened sanctions against Chinese AI companies over allegations they used American models to train competing AI systems.