2 releases covered on ThursdAI · together.ai ↗
Parcae: stable looped transformer matches a model twice its size
Together AI and UCSD researchers introduced Parcae, a stable architecture for looped language models that comes with scaling laws and matches the quality of a transformer twice its size. Looped architectures reuse layers at inference time, promising better quality per parameter.
DeepCoder-14B: open RL-finetuned coder beats DeepSeek R1 and o3-mini on coding
Together AI and Agentica (UC Berkeley Sky Computing Lab) released DeepCoder-14B-Preview, a reasoning model finetuned with RL that beats DeepSeek R1 and even o3-mini on several coding benchmarks. The project aims to democratize RL: the team open-sourced the model, the training dataset, the Weights & Biases logs, and the eval logs. Guest Michael Luo from Agentica joined the show to discuss the release.
Never miss a Together AI & Agentica (UC Berkeley) launch — we cover every release live, every Thursday.