DeepSeek Unveils Janus-Pro and JanusFlow Multimodal Models on Lunar New Year’s Eve
Chinese AI company DeepSeek released two new multimodal frameworks, Janus-Pro and JanusFlow, in the early hours of January 28, coinciding with Lunar New Year’s Eve. Janus-Pro serves as an upgraded version of the original Janus model, featuring a unified architecture for both multimodal understanding and generation. By decoupling visual encoding, the framework enhances adaptability and performance across diverse tasks. Notably, benchmark tests indicate that Janus-Pro outperforms OpenAI’s DALL-E 3 in image generation capabilities. Consistent with DeepSeek’s open-source strategy, the model is available in two parameter sizes: JanusPro 1.5B and JanusPro 7B. This launch follows the recent success of DeepSeek-V3, which topped leaderboards for open-source models, causing significant stock declines for major tech firms like Nvidia and Oracle. The release underscores DeepSeek's growing influence in the global artificial intelligence sector, challenging established Western competitors with high-performance, open-access alternatives. The timing of the announcement during a major cultural holiday highlights the company's aggressive development pace and strategic positioning in the competitive AI landscape.
Wire timeline
DeepSeek Unveils Janus-Pro and JanusFlow Multimodal Models on Lunar New Year’s Eve
Chinese AI company DeepSeek released two new multimodal frameworks, Janus-Pro and JanusFlow, in the early hours of January 28, coinciding with Lunar New Year’s Eve. Janus-Pro serves as an upgraded version of the original Janus model, featuring a unified architecture for both multimodal understanding and generation. By decoupling visual encoding, the framework enhances adaptability and performance across diverse tasks. Notably, benchmark tests indicate that Janus-Pro outperforms OpenAI’s DALL-E 3 in image generation capabilities. Consistent with DeepSeek’s open-source strategy, the model is available in two parameter sizes: JanusPro 1.5B and JanusPro 7B. This launch follows the recent success of DeepSeek-V3, which topped leaderboards for open-source models, causing significant stock declines for major tech firms like Nvidia and Oracle. The release underscores DeepSeek's growing influence in the global artificial intelligence sector, challenging established Western competitors with high-performance, open-access alternatives. The timing of the announcement during a major cultural holiday highlights the company's aggressive development pace and strategic positioning in the competitive AI landscape.
TechNode