Thinking Machine Labs Launches Inkling: Ex-OpenAI CTO Mira Murati's AI Startup Unveils Its First Open-Weight Model

Former OpenAI Chief Technology Officer Mira Murati's artificial intelligence startup, Thinking Machine Labs, has officially unveiled its first in-house AI model, Inkling.

Positioned as an open-weight multimodal model, Inkling has been trained entirely from scratch and is designed to be extensively customized by developers and enterprises. The launch marks the company's first major product since its record-breaking funding round and signals its ambition to challenge established AI players with a more open approach.

Mira Murati Announces Inkling

Sharing the announcement on X, Mira Murati wrote:

"Our first model, Inkling. Trained from scratch, weights are open, fine-tunable on Tinker today."

The model can be customized using Tinker, Thinking Machine Labs' proprietary fine-tuning platform designed for enterprise AI development.

Thinking Machine Labs was founded in February 2025 by Murati and several former OpenAI executives. The startup reportedly raised around $12 billion in seed funding, one of the largest early-stage funding rounds in AI history.

What Is Inkling?

Inkling is a multimodal AI model capable of understanding:

• Text
• Images
• Audio
• Code

Although it currently generates text-based outputs, it can produce:

• Code
• Structured data
• Styled documents

According to Thinking Machine Labs, Inkling features:

• 975 billion total parameters
• 41 billion active parameters during inference
• Training on 45 trillion multimodal tokens

The company says the model was trained entirely from scratch rather than building upon an existing foundation model.

What Makes Inkling Different?

One of Inkling's biggest differentiators is that it is an open-weight model.

Unlike proprietary AI systems such as OpenAI's GPT series or Anthropic's Claude models, developers can:

• Download the model weights
• Run Inkling locally
• Modify and fine-tune it for specific use cases

This approach gives organizations greater control over deployment, privacy and customization while reducing reliance on cloud-based AI services.

Thinking Machine Labs believes organizations achieve better results when AI models are tailored to their own workflows rather than relying on one-size-fits-all systems.

Built for Enterprise Customization

Rather than positioning Inkling as the industry's most powerful general-purpose AI model, Thinking Machine Labs describes it as a highly adaptable foundation for enterprise applications.

Key capabilities include:

• Adjustable reasoning effort to balance speed and accuracy
• Better calibration to acknowledge uncertainty instead of confidently guessing
• Extensive fine-tuning through the Tinker platform

The company says these features make Inkling suitable for businesses that want greater control over AI behaviour.

How Does Inkling Compare With GPT and Claude?

Thinking Machine Labs released benchmark comparisons showing Inkling performs competitively against several leading open-weight models.

According to the company:

• Inkling matched Nvidia Nemotron 3 Ultra on Terminal Bench 2.1 while using significantly fewer tokens.
• It currently trails frontier proprietary models such as GPT-5.6, Claude Fable 5, and Kimi K2.6 on certain benchmarks.

The company acknowledged that Inkling is not the strongest overall model available, but emphasized that its flexibility and openness are its primary strengths.

Inkling-Small Also Announced

Alongside the flagship model, Thinking Machine Labs introduced Inkling-Small, a lighter version featuring:

• 12 billion active parameters
• Lower computational requirements
• Reduced latency
• Lower deployment costs

The company plans to expand the Inkling family with additional models in the future.

AI Strategy Focuses on Openness

Thinking Machine Labs has consistently argued that artificial intelligence should not be controlled by only a handful of companies.

Instead, the startup advocates for decentralized AI development, allowing organizations to own, customize and deploy models independently.

The launch of Inkling reflects that philosophy by giving developers access to model weights while providing enterprise tools for customization.

What's Next for Thinking Machine Labs?

Inkling represents the startup's first public AI product after nearly eighteen months of development.

With around 200 employees and substantial financial backing, Thinking Machine Labs is expected to continue expanding the Inkling ecosystem while competing in the rapidly evolving enterprise AI market.

Whether its open-weight strategy can challenge dominant proprietary AI platforms remains one of the industry's most closely watched questions.

Key Highlights

• Thinking Machine Labs has launched its first AI model, Inkling.
• The startup was founded by former OpenAI CTO Mira Murati.
• Inkling is an open-weight multimodal AI model trained from scratch.
• The model supports extensive customization using the company's Tinker platform.
• Thinking Machine Labs also introduced Inkling-Small for lower-cost deployments.

FAQs

What is Inkling?

Inkling is the first AI model developed by Thinking Machine Labs. It is an open-weight multimodal model trained from scratch for enterprise customization.

Who founded Thinking Machine Labs?

Thinking Machine Labs was founded by former OpenAI CTO Mira Murati along with several former OpenAI executives.

What makes Inkling different from GPT?

Unlike proprietary GPT models, Inkling is open-weight, allowing developers to download, modify and run the model locally.

Can companies customize Inkling?

Yes. Thinking Machine Labs designed Inkling specifically for extensive fine-tuning through its Tinker platform.

Does Inkling outperform GPT and Claude?

The company says Inkling performs competitively against several open-weight models but does not claim it surpasses frontier proprietary models like GPT-5.6 or Claude Fable 5.