RadixArk has introduced DaisyRose-27B, a cutting-edge coding model boasting 27 billion parameters. This model is distinctive as it has been exclusively post-trained using reinforcement learning techniques, building upon the foundation of Qwen3.8-27B.
The performance of DaisyRose-27B is noteworthy, achieving a pass rate of 12.0 on Terminal Bench 4.1, which more than doubles the previous benchmarks. This advancement allows developers to leverage its capabilities while operating on a single GPU, making it a powerful tool for coding tasks.
The release of DaisyRose-27B signals a significant step forward in the realm of AI-driven coding solutions, emphasizing the potential of reinforcement learning in enhancing model performance and usability in real-world applications.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.