Inkling open-weight model launches as Murati targets OpenAI

Thinking Machines has released Inkling, its first general-purpose AI model, giving Mira Murati’s startup a product to show after a year dominated by hiring and funding headlines. The open-weight release tests whether a well-capitalised OpenAI spinout can win enterprise buyers that want more control than a closed API normally gives them.

The company has been watched largely because of who Murati hired and how much capital she raised. WIRED reported that Thinking Machines raised $US2 billion (about $3.1 billion) at a $US12 billion valuation (about $18.6 billion) last year. Inkling changes the question from whether the team is credible to whether its first model is useful enough to compete with OpenAI and Anthropic.

According to the Inkling model card, the flagship model uses a mixture-of-experts design with 975 billion total parameters, 41 billion active parameters and a one million token context window. Thinking Machines said it trained the model on 45 trillion tokens and built it to accept text, image, audio and video inputs. It is also releasing Inkling-Small, a 12 billion-active-parameter variant. Those specifications put the company into sales conversations now led by OpenAI, Anthropic and other suppliers selling managed access to large models, but Inkling’s pitch is as much about deployment as scale.

In its launch post, the company said it wanted users to modify the model rather than treat it as a fixed hosted product:

Today we are advancing our mission by releasing a model we trained from scratch with the full weights available, so that people can make it their own.

Thinking Machines Lab

The company also worked to lower the temperature around the launch. In the same announcement, Thinking Machines said Inkling is “not the strongest overall model available today, open or closed”, an unusual caveat in a market where model releases are often sold on benchmark wins. That points to a bet on adoption, developer trust and deployment flexibility, not only a claim of technical superiority.

Why open weights matter

For enterprise buyers, including Australian teams weighing data handling and governance settings, open weights offer a different trade-off from API-only frontier models. A published model can be inspected, fine-tuned and, in some cases, hosted closer to a customer’s own stack. A system at Inkling’s scale will still push many organisations towards cloud infrastructure and specialist support, but the control story is different from a purely hosted product.

That distinction matters as companies try to move generative AI from internal demos into production systems. For Australian organisations, the appeal is less about chasing the largest benchmark claim than about knowing where a model sits, how it can be tuned and which controls can be wrapped around it before deployment.

Murati’s company is trying to use that gap. Reuters reported that the launch puts the startup into more direct competition with OpenAI and Anthropic, while Axios wrote that Murati is pitching a cheaper and more adaptable alternative to those closed-model leaders. For buyers, the practical question is whether that openness can outweigh the scale, maturity and ecosystem advantages the incumbents already have. Inkling does not look like a benchmark knockout; it gives Thinking Machines a serious product, a point of difference from closed rivals and something enterprise customers can test against their own workflows.