Problem

Thinking Machines Lab addresses the gap between general-purpose AI models and the specialized needs of people and organizations. Its stated goal is to make AI systems more widely understood, customizable, and generally capable, because many real-world problems are not solved well by even the strongest generalist models when they lack an organization’s specialized knowledge.

The practical pain is therefore less about access to an AI model than about adapting one to a particular workflow, dataset, modality, or judgment standard. A prominent use case is forecasting: timestamped data can be converted into verifiable reinforcement-learning tasks that train models to make more accurate and well-calibrated predictions. The economic proposition is to obtain domain-specific performance without each team having to build and operate its own model-training infrastructure, although the available evidence does not quantify customer ROI or savings.

Product / Service

The core delivery model is Tinker, a training API for researchers and developers. It gives users control over model training and fine-tuning while Thinking Machines handles the underlying infrastructure. Tinker supports a range of open-source models, uses usage-based pricing, and charges separately for checkpoint storage, making customization available as a managed software service rather than requiring customers to assemble a full training stack.

Thinking Machines also provides the models and interfaces that make this platform useful. Inkling is its first open-weights model: a multimodal, generalist base designed less to win every benchmark than to be efficiently customized through Tinker. The company has also previewed interactive collaboration systems that combine a real-time, multimodal interaction model with an asynchronous background model for reasoning, tool use, and longer-running work, aiming to deliver both conversational responsiveness and agentic capability. Inkling is available through Tinker and through APIs from several model-hosting partners, with its weights published on Hugging Face.

Market

Thinking Machines Lab competes in the frontier-AI and AI developer-infrastructure markets, with a differentiated emphasis on customizable multimodal models and human-AI collaboration rather than a single one-size-fits-all assistant. Its closest high-profile strategic comparisons are OpenAI and Anthropic, while its open-weights models and managed fine-tuning platform also place it alongside the broader open-model and model-hosting ecosystem.

The company has substantial financing and early product traction but limited publicly established commercial traction in the evidence reviewed. Reuters reported a $2 billion funding round at a $12 billion valuation in July 2025, and later reported Nvidia as an investor; Tinker launched publicly in October 2025, and Inkling launched in July 2026 with distribution through Together AI, Fireworks, Modal, Databricks, and Baseten. A January 2026 company report described Thinking Machines Lab as pre-revenue, and the available evidence does not establish that revenue or a customer count had subsequently been disclosed by August 1, 2026.