DeepSeek V4 Flash Vision Exp is here, and it just gave DeepSeek's most efficient model actual eyes.
After staying text-only for years, deepseek_ai quietly dropped a vision-enabled version of V4 Flash into their API, and this breakdown covers everything that actually matters
I finally understand why Deepseek is so stupid.
Today I tested the DeepSeek-V4-Flash-Vision-Exp visual model. If it does not assist in online search, this capability is still a bit poor.
A simple lawn for cats and dogs is still identifiable. But if it is a character, the corresponding content needs to be verified together with online searches.
Found this: Free Claude Code.
Its Open-source, MIT licensed, ~46.5K stars. Lets you run Claude Code with models from OpenAI, Google, DeepSeek, and more, through a proxy. Repo Link -
You have reached the end of the archive
All of deepseek