Thinking Machines Lab Drops Its First Model

What changed

Thinking Machines Lab launched Inkling, an open source AI model with 975 billion parameters. Unlike many text-only models, Inkling is trained to process both video and audio inputs. This positions it in a distinct niche where multimodal understanding is critical, setting it apart from competitors that primarily focus on language alone. As an open source release, Inkling aims to attract builders who require a large-scale, multimodal model without the usual closed access restrictions.

Why builders should care

Multimodal AI is becoming essential for tasks that combine sound and visuals—think video editing, surveillance, accessibility tools, and interactive agents. Inkling’s training on both video and audio reduces the friction for developers needing one model to handle diverse input types. Being open source not only cuts costs for exploration but also provides greater flexibility to customize or build on top of a large-scale multimodal framework. This can accelerate innovation in areas where existing models, like those from Anthropic or OpenAI, may lock users into API limits or lack native multimodal training.

The practical takeaway

Inkling’s combination of scale and multimodal capability will pressure established players to offer more flexible or specialized tools for audio-visual tasks. Builders working with video and sound data now have a potent, free alternative that can be integrated without licensing hurdles. This could lower entry barriers for startups or researchers testing new concepts involving real-time video comprehension or audio-visual agent design. At the same time, deploying such a large model requires significant compute resources, so operational costs remain a factor.

What to watch next

The key next step is seeing how Thinking Machines Lab supports Inkling’s adoption—whether through developer tools, documentation, or partnerships. Watch for new use cases emerging from the open source community that push multimodal AI beyond prototype stages. Also, observe how competitors respond: will Anthropic and OpenAI enhance their multimodal offerings or adjust pricing to retain relevance? Finally, monitor infrastructure providers to see if they start offering optimized environments to run giant models like Inkling efficiently.

AI Quick Briefs Editorial Desk