Meet Inkling: 1st AI model from Mira Murati's start-up
What's the story
Thinking Machines Lab, an artificial intelligence (AI) startup co-founded by former OpenAI CTO Mira Murati, has launched its first in-house AI model named Inkling. Unlike the flagship models from OpenAI, Anthropic, or Google, Inkling is an open-weight system. This means that developers and companies can download and modify it directly. The move is a major departure from the proprietary nature of most leading AI models today.
Model details
Inkling is a massive multimodal model
Inkling is a mixture-of-experts system with a whopping 975 billion parameters. However, it only uses about 41 billion for any given task.
The model was trained on an enormous dataset of 45 trillion tokens comprising text, image, audio, and video data.
It can reason natively across all four modalities but currently only produces text outputs such as code, styled artifacts, and structured data.
Performance metrics
Inkling can flag uncertainty instead of guessing
Inkling is designed to provide calibrated answers, including flagging uncertainty instead of guessing. It also allows users to adjust "thinking effort" for speed.
The company claims that Inkling uses a third as many tokens as NVIDIA's latest open-weight model, Nemotron 3 Ultra, to achieve the same coding performance.
This shows how efficient and adaptable the new AI model can be in different scenarios.
Market strategy
Thinking Machines offers Tinker for customizations
Thinking Machines is marketing Inkling not as a finished product but as a starting point for organizations to fine-tune themselves through Tinker, the company's model-customization platform.
This means customers are responsible for ensuring their customizations are safe.
The approach is different from other companies like OpenAI, Anthropic, and Google, which built general-purpose chatbots first with autonomous features added later.
Customization focus
Microsoft CEO warns enterprises using proprietary AI models
Thinking Machines believes that AI trained centrally by one company and then set in stone underperforms AI that organizations shape themselves.
Microsoft CEO Satya Nadella, whose company has invested billions in OpenAI and Anthropic, also warned enterprises using proprietary AI models pay twice: once in subscription costs and again by handing over business knowledge embedded in their prompts and corrections.
Growth strategy
Thinking Machines has been tight-lipped about its spending
Thinking Machines has been tight-lipped about its spending, having partnered with NVIDIA in March to deploy a gigawatt of Vera Rubin computing capacity.
The company trained Inkling entirely on NVIDIA's GB300 NVL72 systems but hasn't revealed how it plans to cover those costs.
Despite the lack of transparency around funding and financials, the company claims it took just nine months to bring its tech to market, a feat that took OpenAI five years and Anthropic three.
Twitter Post
Official announcement
Today, we are introducing Inkling.
— Thinking Machines (@thinkymachines) July 15, 2026
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.https://t.co/Ghebq5mG30
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵