Thinking Machine Labs, the artificial intelligence startup founded by former OpenAI Chief Technology Officer Mira Murati, has officially unveiled its first in-house AI model, Inkling. The launch marks a major milestone for the company, which has spent more than a year building its AI platform after emerging as one of the world’s most closely watched startups.

Unlike many leading AI models that operate exclusively through cloud services, Inkling is an open-weight model, allowing developers and businesses to download, modify and fine-tune it for their own requirements.

Inkling Brings Open-Weight AI to Developers

Thinking Machine Labs says Inkling was trained entirely from scratch and supports multimodal understanding, enabling it to process text, images, audio and code.

Although the model currently generates only text-based outputs, it can produce code, structured data and formatted content while reasoning across multiple input types.

The company describes Inkling as a highly customisable AI model rather than one built solely to top industry benchmark rankings.

Built for Enterprise Customisation

According to the company, Inkling contains 975 billion total parameters, with 41 billion active parameters used during inference for a specific task.

The model was trained using approximately 45 trillion tokens, covering text, images, audio and video data.

Thinking Machine Labs says its primary goal is to provide organisations with an AI system that can be adapted to specific business workflows instead of relying on a one-size-fits-all approach.

Developers can further customise Inkling using Tinker, the company’s own AI fine-tuning platform.

How Inkling Differs From Closed AI Models

Unlike proprietary models from companies such as OpenAI and Anthropic, Inkling follows an open-weight approach.

This enables organisations to:

• Run the model locally
• Fine-tune it for specialised tasks
• Reduce dependence on cloud-based AI services
• Lower operational costs
• Improve privacy and security

The company believes businesses benefit more when they have greater control over how AI systems learn and operate.

Focus on Reliable AI Responses

Thinking Machine Labs says Inkling has been designed to produce more calibrated responses.

Instead of confidently generating uncertain information, the model attempts to indicate uncertainty whenever appropriate.

Users can also adjust the model’s “thinking effort,” allowing them to balance response speed with reasoning quality depending on their workload.

Benchmark Performance

The company shared benchmark results comparing Inkling with several leading AI models.

On the Terminal Bench 2.1 benchmark, Inkling reportedly delivered performance comparable to Nvidia’s Nemotron 3 Ultra while consuming significantly fewer tokens.

However, the company acknowledged that flagship models such as GPT 5.6, Claude Fable 5, and Kimi K2.6 continue to outperform Inkling on several overall benchmarks.

Rather than competing solely on raw performance, Thinking Machine Labs says its focus is flexibility, efficiency and enterprise customisation.

Trained on Nvidia Infrastructure

Thinking Machine Labs confirmed that Inkling was trained entirely on Nvidia GB300 NVL72 systems through a strategic partnership announced earlier this year.

While the model’s pre-training process was completed from scratch, the company said some early post-training data included outputs from open-weight models such as Kimi K2.5 before reinforcement learning became the primary training method.

Inkling-Small Also Announced

Alongside its flagship release, the startup introduced Inkling-Small, a lighter version featuring 12 billion active parameters.

The smaller model is designed to deliver faster performance with lower computing costs, making it suitable for developers and organisations seeking efficient AI deployment.

Thinking Machine Labs also confirmed that additional models in the Inkling family are already under development.

A Vision for Decentralised AI

The release reflects the company’s broader philosophy that artificial intelligence should not remain concentrated within a handful of technology companies.

Thinking Machine Labs believes organisations should have greater ownership of the AI systems they use, enabling deeper customisation, stronger privacy and lower long-term costs.

The startup has grown rapidly since its launch and now employs around 200 people while positioning itself as a major challenger in the rapidly evolving AI industry.

What Comes Next?

The launch of Inkling marks Thinking Machine Labs’ first major public product since the company was founded. With its open-weight architecture, enterprise-focused design and emphasis on customisation, the startup is aiming to carve out a unique position in the increasingly competitive AI market.

As demand grows for flexible and cost-effective AI solutions, Inkling could become an attractive option for developers and businesses seeking alternatives to proprietary AI platforms.

FAQs

1. What is Inkling?

Inkling is the first AI model developed by Thinking Machine Labs, the startup founded by former OpenAI CTO Mira Murati.

2. Is Inkling an open-source AI model?

Inkling is an open-weight AI model, meaning developers can download, modify and fine-tune its model weights, although it is not fully open source.

3. What can Inkling do?

The model can understand text, images, audio and code while generating text-based outputs, including code and structured data.

4. Who founded Thinking Machine Labs?

Thinking Machine Labs was founded by former OpenAI Chief Technology Officer Mira Murati along with several former AI industry executives.

5. How large is the Inkling model?

Inkling has 975 billion total parameters with 41 billion active parameters used for each task.

6. What is Inkling-Small?

Inkling-Small is a lightweight version of the model featuring 12 billion active parameters, designed for faster and lower-cost deployment.

7. How is Inkling different from ChatGPT or Claude?

Unlike proprietary AI models, Inkling uses an open-weight architecture that allows businesses to run and customise the model locally.

8. Why is Inkling important?

Inkling represents Thinking Machine Labs’ entry into the AI market and offers developers an enterprise-focused, highly customisable alternative to closed commercial AI models.