Mind Lab

Mind Lab is an AI research lab focused on continual learning and LLM post-training. It builds AI models that keep improving after deployment and turns enterprise expertise into proprietary model capabilities. Its research covers open continual learning, long-context reinforcement learning, efficient online memory for large language models, and parameter-efficient fine-tuning at scale. Papers, models and code are released openly on Hugging Face and GitHub.

For enterprise AI, Mind Lab runs Mint Recursive, an industry-focused platform for model training and inference. Teams in pharmaceuticals, finance, the public sector and gaming get an end-to-end service that starts with a dedicated benchmark built from their real workflows, then moves through data development, post-training, inference and deployment. The result is a vertical AI model that reasons the way a company’s own specialists do, from evidence review in drug R&D to medical chart quality control.

Mind Lab post-trains on frontier open base models, including Qwen, GLM, DeepSeek, Kimi and MiniMax. It handles everything from small task-specific models to trillion-parameter flagships. With LoRA, shared bases and distributed optimization, it has cut GPU cost by more than 90% on shipped projects, and specialist training starts at $1,000. Models can run in private, VPC or hybrid environments.

Companies can work with Mind Lab in three ways. They can partner with its forward-deployed experts, use its automated training platform, or bring their own loss functions and RL loops onto Mind Lab’s training and inference infrastructure. Mind Lab is also the research team behind Macaron, a personal AI agent.