What Thinking Machines Lab does
Thinking Machines Lab is an AI research and product company focused on building customizable frontier systems and “open-weights” models, alongside tools that let developers and researchers fine-tune those models in controlled ways. The company’s product strategy centers on (1) releasing frontier-scale open models (e.g., Inkling and Inkling-Small) and (2) providing Tinker, a managed training API for LoRA-based fine-tuning that abstracts away distributed-training infrastructure while preserving control over the training loop and algorithms. In parallel, Thinking Machines is pursuing “interaction models” as a model-level approach to real-time, multimodal human-AI collaboration, aiming to make interactivity a property of the model rather than only something added by external harnesses.
Business-wise, Thinking Machines appears positioned to serve three overlapping user groups: (a) researchers and builders who want transparent, inspectable model weights and training recipes; (b) developers who need to tailor frontier open models to domain-specific data and objectives via LoRA fine-tuning; and (c) enterprises and institutions that want scalable customization and deployment workflows without being forced into fully closed model ecosystems. Its technical differentiation emphasizes multimodality (text, audio, and vision), long-context capability (up to 1M tokens for Inkling-family), and training pragmatics such as reward modeling/training stability for fine-tuned outcomes. The company also explicitly addresses safety and misuse risk as part of its open-weights release strategy and offers grant programs to support safety and interactivity research.