Researchers trained a 1.14 billion parameter Learner 1.0 model on ten unrelated skills sequentially, one example at a time with no replay, demonstrating that later skills did not erase earlier ones. Final evaluation showed minimal performance degradation across retained skills, with most skills maintaining their learned scores after all ten sequential trainings.
Thomson-1.0-Small is an open-weight frontier foundation model developed by Thomson Reuters using continual learning on the Qwen3.6-35B base model. It achieves high performance across legal, tax, and journalism domains through constitutional value alignment, data-centric training on 19T+ tokens, and agentic deep research capabilities, demonstrating that frontier model performance is achievable by institutions beyond heavily funded players.
Mini-AGI is a continual learning language model that trains from scratch on 8GB VRAM by storing weights on disk and paging them as needed. It learns continuously from a data stream without catastrophic forgetting, enabling users to train and own their own models on consumer hardware.
Researchers introduce long-horizon memorization, a challenge where language models must learn 100 tasks through continual fine-tuning without catastrophic forgetting. They show that composing multiple continual learning mechanisms—combining data, function, and weight anchors with merged LoRA—improves retention from 1.2% to 34.9%, a 28-fold improvement over naive sequential fine-tuning.