Enterprise Model Forking Platform

The key strategic move is that Thinking Machines is trying to sell control, not just model access. Instead of giving a company a chat window over a centrally managed model, it is building an open weights base model and a training stack that lets customers change the model itself for their own domain, data, and policies. That makes the product look less like ChatGPT, and more like enterprise software for owning an internal AI system.1 2 3 4 5

• Forking here means more than prompts or retrieval. Tinker is built around fine tuning and post training on customer data, with support for open weight models and direct weight adaptation. That gives a hospital, bank, or manufacturer a way to shape how the model reasons and responds before it ever reaches end users.1 2 3 6
• That is a sharper enterprise pitch than ChatGPT. OpenAI offers customization and fine tuning, but ChatGPT Enterprise is still primarily a hosted application with workspace controls and model access. Thinking Machines is positioning its stack closer to an internal model factory, where the customer can train, serve, and evolve a version that is more their own.4 5 2 3
• The closest comparables are the open model and infrastructure layer, not consumer chat apps. Fireworks, Hugging Face style open model workflows, and other open weight ecosystems win when developers want speed, lower lock in, and deeper model level control. Thinking Machines is pushing that pattern upmarket with frontier grade multimodal models and managed training infrastructure.2 3 6 7

If this works, the market splits more clearly in two. Closed assistants stay strong for general purpose chat, while customizable base models become the default for regulated and specialized work. That would put Thinking Machines in the part of the stack where enterprise budgets are larger, switching costs are higher, and customer data makes each deployed model more defensible over time.1 2 3 4 6