Mind Lab Opens Mint Recursive Post-Training to Enterprises Building Industry-Specific AI Models

Mind Lab's Mint Recursive covers benchmark design, data, post-training, inference and deployment in one workflow, with specialist training starting at $1,000

San Francisco, CA (PRUnderground) October 21st, 2026

Mind Lab (mindlab.im), the AI research lab focus on continue learning and posttrain, is making its Mint Recursive platform available to enterprises that need AI models trained for their own industry. Mint Recursive provides enterprise model post-training that turns a company's expert experience into model capabilities the company owns, for organizations across industries such as pharmaceuticals, finance, the public sector and gaming.

Starting from a shared benchmark

A general-purpose model is a starting point, not a finished product. Mind Lab begins each engagement by building a benchmark with the customer, one that defines the task goal, what correct behavior looks like, the mistakes that come up most often and the acceptance line a model has to clear. Data development, LLM post-training and deployment then run in a single loop, so each iteration is measured against a standard the customer helped set.

Lower costs and three ways to work

Cost is often what keeps teams from training their own models. Mind Lab combines LoRA post-training, shared base models and distributed optimization, so fewer parameters are updated and less compute is spent.

The company reports that GPU costs on shipped projects have fallen by more than 90%, and specialist training starts at $1,000. Customers can choose base models from the GLM, DeepSeek, Kimi, Qwen and MiniMax families and deploy in private, VPC or hybrid environments.

Customers also decide how much of their own team to commit. Mind Lab's forward-deployed engineers (FDEs) can run the benchmark, data, post-training and deployment end to end. Teams that define their own model, data and configuration can hand scheduling, training and delivery to the platform's automated training. Teams that want full control can write their own loss functions and training loops, while the lab provides clusters, trajectory sampling, weight sync and inference.

Research behind the platform

The platform grows out of the lab's work in reinforcement learning. Mind Lab was among the early teams to run end-to-end LoRA reinforcement learning on trillion-parameter mixture-of-experts (MoE) models. When the lab introduced Mint Recursive in September, it also released Macaron-V1.1, a 752B-parameter agent model post-trained from GLM-5.3, with open weights. Moving from Macaron-V1 to Macaron-V1.1 took two weeks on the platform. Across 60 UI tasks, first-attempt delivery success rose from 71.67% for GLM-5.3 to 96.67% for Macaron-V1.1.

In healthcare and life sciences, for instance, Mind Lab supports literature retrieval and evidence review, drug discovery and R&D intelligence, and medical chart quality control. Its shipped work in this area covers 27 quality-control indicators and more than 8,000 physician-experience traces.

"An enterprise's most valuable training signal is the judgment of its own experts," said Chen Kaijie, Founder of Mind Lab. "That judgment lives in the reviews, corrections and decisions a general model never sees. Post-training is how it becomes part of a custom AI model the company owns. It doesn't need to start as a big program. It can start with one real task."

Enterprises ready to build proprietary AI models can start with one real task and contact Mind Lab to define the benchmark for it.

About Mind Lab

Mind Lab is an AI research lab focused on LLM post-training. Its research centers on experiential intelligence, the idea that models should keep learning from real-world experience. The lab developed Mint Recursive, a platform for post-training and inference. It also developed the Macaron series of agent models and has released Macaron model weights as open source on Hugging Face. Learn more at https://www.mindlab.im/ .

Media Contact

Mind Lab

[email protected]

Original Press Release.