Thinking Machines Lab, the artificial intelligence startup co-founded by former OpenAI CTO Mira Murati, has introduced its inaugural in-house AI model, named Inkling. Launched this week, Inkling distinguishes itself from flagship models offered by industry leaders like OpenAI, Anthropic, or Google by being open-weight. This crucial design choice allows external developers and companies to download and directly modify the model, fostering a new paradigm for AI deployment.

Inkling operates as a mixture-of-experts system, boasting 975 billion total parameters, though it efficiently utilizes only a fraction—approximately 41 billion—for any given task. This architecture ensures faster and more cost-effective operation for very large models. The model underwent training on an extensive dataset of 45 trillion tokens encompassing text, image, audio, and video, enabling native reasoning across all four modalities. Currently, its outputs are primarily text-based, including code, styled artifacts, and structured data. Thinking Machines highlights Inkling’s ability to provide calibrated answers, signaling uncertainty rather than making guesses, and allowing users to adjust “thinking effort” to balance speed and precision. The company claims Inkling can achieve comparable coding performance to Nvidia’s Nemotron 3 Ultra, its latest open-weight model, using a third fewer tokens.

Thinking Machines is not positioning Inkling as the strongest overall model available today, whether open or closed. Instead, its strategic focus is on well-rounded performance and, critically, customizability. The company markets Inkling less as a finished product and more as a foundational tool for organizations to fine-tune via Tinker, its dedicated model-customization platform. This approach contrasts sharply with the general-purpose chatbot strategies of OpenAI’s ChatGPT, Anthropic’s Claude, and Google’s Gemini, which prioritize broad utility and autonomous features. The responsibility for ensuring the safety of these customized applications, for instance, rests with the customers.

The industry is seeing a growing consensus around the limitations of proprietary, one-size-fits-all AI. Microsoft CEO Satya Nadella has cautioned that enterprises using closed models effectively incur double costs: subscription fees and the implicit transfer of valuable business knowledge embedded in their prompts. Similarly, Hugging Face CEO Clem Delangue predicts a shift where frontier models are reserved for experimentation, while most production AI work moves to private or open-source alternatives—a vision Thinking Machines is actively building towards. A project with Bridgewater Associates, the world’s largest hedge fund, demonstrated the potential of this approach, with a fine-tuned open-source model achieving 84.7% on financial reasoning tests, outperforming top proprietary models at a fraction of the cost. Thinking Machines also notes its rapid development, achieving market readiness in approximately nine months, significantly faster than some larger competitors.

Key Takeaways

• Thinking Machines Lab, co-founded by Mira Murati, has launched Inkling, its first open-weight AI model designed for enterprise customization.
• Inkling is positioned as a foundational tool for organizations to fine-tune for specific needs, challenging the general-purpose, 'one-size-fits-all' approach of major AI labs.
• The model emphasizes calibrated answers, efficiency, and multimodal reasoning, with revenue expected from its customization platform, Tinker, rather than direct model usage fees.

Editor’s Analysis & Impact

The launch of Inkling by Thinking Machines Lab signals a significant strategic shift within the AI industry, moving beyond the dominance of general-purpose models. By championing an open-weight, customizable approach, Thinking Machines is betting on the power of specialized AI tailored to specific enterprise needs. This could democratize access to advanced AI, allowing businesses to integrate sophisticated capabilities without proprietary lock-in or the high costs associated with large, closed models. The success of Inkling and its Tinker platform will hinge on enterprises’ willingness to invest in fine-tuning and the availability of skilled AI talent. If this model proves effective, it could foster a more diverse and competitive AI ecosystem, where efficiency and domain-specific expertise become as crucial as raw computational power, potentially reshaping how businesses adopt and leverage artificial intelligence.

Frequently Asked Questions

Q: What is Inkling and who developed it?
A: Inkling is the first in-house AI model released by Thinking Machines Lab, an AI startup co-founded by former OpenAI CTO Mira Murati. It is an open-weight model, meaning developers and companies can download and modify it directly for their specific needs.

Q: How does Inkling's approach differ from other major AI models like ChatGPT or Gemini?
A: Unlike general-purpose chatbots such as ChatGPT, Claude, or Gemini, Inkling is designed as a customizable 'starting point' for enterprises. It focuses on providing calibrated answers and allows organizations to fine-tune it through Thinking Machines' Tinker platform, rather than being a ready-to-use, one-size-fits-all solution.

Q: What is the core philosophy behind Thinking Machines' strategy with Inkling?
A: Thinking Machines believes that AI models organizations can adapt and specialize for themselves will ultimately outperform generic, centrally trained models. This approach aims to leverage specific enterprise expertise, offer more efficient, tailored solutions, and provide a more flexible alternative to proprietary AI systems.