Mind Lab, the company behind the personal agent app Macaron, has released the full version of its Macaron-V1 model on July 21. Founded in October 2025 with more than 30 employees, Mind Lab’s team largely comes from xAI, DeepMind, DeepSeek, ByteDance Seed, MIT, and Tsinghua University. Its direction aligns with the continual learning approach championed by Turing Award winner Richard Sutton and OpenAI’s former CTO Mira Murati. The company released the Macaron-V1-Preview model in June, and just two weeks after commercialization began, its annual recurring revenue reached USD 10 million. The preview was built by attaching five LoRA expert modules to GLM-5.1, each with about one billion parameters, and outperformed its base model as well as GPT-5.4 and Claude Opus 4.6 on benchmarks. Mind Lab’s post-training approach uses a mixture of LoRA adapters, allowing the system to dynamically switch expert modules and distill user data into continually updated adapters. The full Macaron-V1 release includes two models, with the flagship version named Venti. According to company-published benchmarks, Macaron-V1 achieved state-of-the-art results in six of 12 tests, with remaining scores close to frontier models.