Thinking Machines Lab, the AI research and products company founded by former OpenAI CTO Mira Murati, has released its first public model, Inkling, a large-scale, open-weights system built for developers and enterprises who want to customise AI rather than rent it through a closed API.
The launch marks the company’s first major product reveal since it emerged from roughly 18 months of low-profile infrastructure work and research. Inkling is available today on Tinker, Thinking Machines Lab’s fine-tuning platform, at two context lengths — 64K and 256K tokens — and is being offered at a 50% discount for a limited time.
What Is Inkling?
Inkling is a mixture-of-experts (MoE) transformer trained from scratch, with 975 billion total parameters and roughly 41 billion active on any given task. That MoE design allows the model to carry the capacity of a very large system while keeping inference faster and cheaper than running all 975 billion parameters at once.
The model is natively multimodal, taking in text, images and audio, and it gives users controllable “thinking effort,” letting developers trade off cost and performance depending on the complexity of a given task.
According to Thinking Machines Lab, Inkling was trained to be a broad, balanced foundation model rather than a benchmark leader. The company has been candid that Inkling is not the strongest overall model on the market today, open or closed, its value proposition instead centres on being a flexible, well-rounded base for customisation, with particular strength in agentic coding and tool use.
Open Weights, Built for Customisation
Unlike the flagship, closed models offered by labs such as OpenAI, Anthropic and Google, Inkling ships as an open-weights model. Developers and companies can download it, fine-tune it on their own data, and deploy it on their own infrastructure, rather than routing every request through a centralised API.
That approach aligns with Thinking Machines Lab’s broader thesis: that AI systems organisations can adapt for their own needs will outperform one-size-fits-all models over time. Inkling is designed to run inside a range of coding and agent harnesses, with training that randomised tool sets and schemas to reduce sensitivity to any one setup.
Getting Started With Inkling
Developers can start building with Inkling today via the Inkling cookbook on Tinker, which walks through fine-tuning and deployment at both supported context lengths. Thinking Machines Lab has framed the release as the first entry in an ongoing model family, with further iterations expected as the company builds out its product line.
Founded in February 2025, Thinking Machines Lab has drawn significant backing from investors, including a landmark seed round that valued the company in the billions, positioning Inkling’s release as one of the most closely watched open-weights launches of 2026.
About Thinking Machines Lab
Thinking Machines Lab is an AI research and products company founded in February 2025 by Mira Murati, former Chief Technology Officer of OpenAI, where she also served as interim CEO during the company’s board upheaval in late 2023. Murati left OpenAI in September 2024 to start the company. Thinking Machines Lab has assembled backing from a roster of prominent technology investors and built out Tinker, its fine-tuning platform, alongside its research work, positioning the company around a central bet: that AI systems organisations can adapt and customise for themselves will outperform one-size-fits-all models sold through closed APIs. Inkling, released in July 2026, is the company’s first public model release.
Sources
- Introducing Inkling — Thinking Machines Lab
- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling — TechCrunch
- Inkling model from Thinking Machines Lab now on Databricks — Databricks Blog
