JD Open-Sources JoyAI-LLM Flash Instruct Version, Adapts to "Lobster" to Empower Developers

For developers and small to medium enterprises, finding an efficient and cost-effective AI model base is a daily pain point. JD recently revealed its AI R&D progress, first open-sourcing the Instruct version of the basic large model JoyAI-LLM Flash, breaking the dilemma of parameter involution.

The model uses the FiberPO reinforcement learning method based on the "fiber bundle" theory, performing excellently in 19 authoritative tests. It has core advantages like efficient response, lightweight deployment, and high accuracy, adapting to multiple fields such as code development, agent building, and terminal applications.

Notably, it precisely adapts to the recently popular JD Cloud "Lobster" series, providing developers with a cost-effective model base. Statistics show that the token call volume of the "Lobster" series increased by 455% month-on-month in the past week.