Amber is a fully open, research-oriented model from LLM360, notable for being released alongside its complete training data and intermediate checkpoints — making it unusually transparent for study and reproducibility. It operates with a modest 2048-token context window, which limits its usefulness for longer documents or extended conversations.