DeepSeek
DeepSeek unveils V4.1-Flash, a lighter model that also understands images
🧒 Discovery
DeepSeek built a super-fast little robot that can now look at pictures as well as read text, like a friend who learned to draw as well as read!
📰 Essential
DeepSeek launched V4.1-Flash, the smallest model in its new architecture family, with native image understanding built in. The Chinese company says it delivers faster inference and lower prices than its V4-Pro model.
🔬 Expert
Since September 14, all requests sent to V4-Pro are automatically rerouted to V4.1-Flash, billed at the latter's lower rate, a way for DeepSeek to shift its user base to the newest architecture without breaking the API. The move underscores the pricing pressure Chinese open-source models keep putting on proprietary Western offerings.