Inkling: the first model of Thinking Machines Lab

Meet Inkling, Thinking Machines Lab's first model: open source model of 975B parameters that understands video and audio. Will it compete with Anthropic and OpenAI?

jueves, 16 de julio de 2026 • 5 min read • Q2BSTUDIO Team

Open source model that understands video and audio

The artificial intelligence ecosystem has received a new contender that promises to redefine the rules of the game. Thinking Machines Lab, a startup that until recently operated in a more discreet profile, has presented Inkling, its first foundational model. With 975 billion parameters, Inkling not only competes in scale with the giants of the sector, but does so from a radically opposite philosophy: it is completely open source. This launch marks a before and after in the way we understand the development of multimodal models, since Inkling was trained to simultaneously understand video, audio and text, opening a range of possibilities ranging from the automation of complex processes to the creation of contextual virtual assistants.

To put it in perspective, most closed models like those from Anthropic or OpenAI have opted for proprietary approaches, limiting access to their architectures and weights. Inkling, on the other hand, is committed to transparency and community collaboration. Not only does this lower the barriers to entry for companies that want to customize AI to their specific needs, but it also fosters an ecosystem of auditing and continuous improvement. Inkling's ability to process multiple sensory modalities makes it an ideal tool for industries such as audiovisual production, intelligent surveillance with advanced cybersecurity, or customer service assisted by AI agents who understand emotions in tone of voice and body language.

From a technical perspective, training a model of 975 billion parameters requires a high-performance cloud infrastructure. Thinking Machines Lab has had to resort to massive clusters, likely combining AWS and Azure cloud services to handle the distributed workload. This is no small detail for companies that want to adopt similar technologies: the choice of cloud provider can make the difference between a successful project and a cost nightmare. In this sense, having a technological ally that understands both the infrastructure and application layers is crucial. That's where Q2BSTUDIO's expertise comes in, offering AI for business integrated with cloud services, allowing organizations of any size to replicate models like Inkling without the need for a top-notch engineering team.

The real differential value of Inkling lies not only in its size, but in its multimodal architecture. The ability to align video and audio representations within the same latent space allows tasks such as the automatic generation of subtitles with emotional context, the classification of scenes without human intervention, or even the detection of anomalies in real time for security systems. The latter connects directly to current cybersecurity needs, where AI models can analyze live video streams to identify suspicious behavior, complementing traditional cybersecurity and pentesting solutions that Q2BSTUDIO implemented for its customers.

However, implementing such a model in an enterprise environment is not trivial. An integration strategy is required that includes fine-tuning with proprietary data, optimizing inference performance, and creating interfaces that allow non-technical users to interact with AI. This is where the custom software solutions and custom applications you develop Q2BSTUDIO become the necessary bridge. For example, a logistics company could train Inkling to recognize packages in surveillance videos and automatically trigger alerts in a management system, all accessible from a custom cross-platform application. The flexibility of open source allows the model to adapt to specific verticals, from healthcare to retail, as long as you have the right partner to materialize it.

Another interesting aspect is the application of Inkling in the field of business intelligence. By being able to process video and audio, it can extract unstructured information from recorded meetings, video surveillance, or user-generated content and transform it into actionable data. Integrated with tools like Power BI, a multimodal model can tag emotions in customer calls or identify visual trends in product catalogs. Q2BSTUDIO, through its business intelligence services and Power BI, helps companies build dashboards that consume these outputs directly, closing the loop from sensory perception to strategic decision-making.

The launch of Inkling also raises questions about AI governance. Being open source, anyone can download the weights, but they can also modify them for malicious purposes. This underscores the importance of an ethical approach and good practices in cybersecurity. Companies that choose to adopt this model must implement additional layers of protection, such as application firewalls, monitoring for anomalous behavior in the models, and periodic audits. Q2BSTUDIO complements its offer with cybersecurity services that assess these risks, offering specific pentesting for AI-based systems and hardening recommendations in cloud infrastructure.

On the competitive level, Thinking Machines Lab seeks to position itself as an open alternative to closed giants. While Claude from Anthropic stands out for its alignment and security, and GPT-4 from OpenAI for its conversational versatility, Inkling is committed to native multimodality and open access. Companies that want to differentiate themselves will need to consider which approach best aligns with their goals. For those who prefer to have full control over the model and its data, Inkling represents an almost unique option. However, managing a 975 billion parameter model requires a technological maturity that few organizations possess. This is where Q2BSTUDIO's consultancy, which specialises in process automation and AI, can guide teams in selecting the right infrastructure, sizing resources and implementing efficient data pipelines.

In addition, Inkling's open source ecosystem encourages the creation of autonomous AI agents capable of interacting with the physical world through video and audio inputs. Imagine a warehouse assistant that receives spoken instructions, identifies objects in real time using cameras, and performs actions such as moving inventory or reporting incidents. This type of solution, based on foundational models adapted with customized applications, is increasingly within reach of companies thanks to the maturity of cloud platforms. Q2BSTUDIO offers integration with AWS and Azure cloud services, ensuring latency and cost are manageable even in high-volume deployments.

Finally, it is important to note that Inkling is still in its early stages of adoption. Thinking Machines Lab has published the weights and architecture, but the documentation and fine-tuning tools are under development. This represents an opportunity for pioneering companies to collaborate in improving the model and, at the same time, benefit from early access. Those that decide to take the step can count on the support of Q2BSTUDIO to design the strategy, from the choice of hardware (local or cloud) to the creation of APIs that expose the capabilities of the model to their legacy systems. Ultimately, Inkling isn't just a model – it's a symbol of where open AI is headed, and companies that know how to leverage it will be better positioned to lead the next wave of innovation.

OUR SERVICES

How we can help you

Do you have a project in mind?

Tell us your vision and we'll turn it into a software solution. Whatever the scope, we make your idea real.