Thinking Machines has released Inkling, a new open-weight model built for customisation. For Australian small businesses, the signal is clear: the next AI advantage is owning the workflow, not just renting the chatbot.
Listen to this article
0:00 / --:--Narrated by Margot Ellis
On 15 July 2026, Thinking Machines released Inkling, its first open-weight foundation model. The announcement is technical on the surface: 975 billion total parameters, 41 billion active, multimodal training across text, images, audio and video, and fine-tuning through the company's Tinker platform. The business signal is simpler. AI is moving from a rented assistant to something a company can shape around its own judgement.
That matters because the first wave of small-business AI adoption has been mostly access. Owners bought a subscription, pasted in a task, tested a few prompts and hoped the tool would understand the business well enough. Sometimes it did. Often it sounded impressive while missing the details that make a quote, product page, customer reply or internal decision actually fit.
Open-weight models point to the next wave. The model is not only something you ask. It becomes a base that can be adapted, governed and wrapped inside a workflow. For a small business, that does not mean running a research lab or fine-tuning a model in the back office. It means the market is shifting towards AI systems that can be made more specific, more accountable and more connected to the business they serve.
Inkling is not pitched as the strongest model in the world. Thinking Machines says the point is a broad, balanced open-weight base for customisation, with multimodal capability and controllable thinking effort. That is an important distinction. For many business workflows, the winner is not the model with the flashiest benchmark. The winner is the system that fits the job, the risk, the data and the customer moment.
A cafe, trade business, clinic, retailer or professional service firm does not need a model that can win every abstract test. It needs an AI layer that understands local offers, service boundaries, tone of voice, booking rules, stock realities, approval limits and the difference between a helpful suggestion and a promise the business must honour. Customisation is where generic AI starts becoming operational AI.
The dangerous reading of open-weight AI is that every owner should start collecting model files and experimenting with infrastructure. That is not the lesson. Most small businesses should not be choosing model weights, managing hosting, tuning datasets or maintaining safety tests themselves. The practical lesson is that AI strategy is becoming more like systems design. The value sits in the operating layer around the model.
We made the same point in our recent piece on AI spend needing an operator. Tools are easy to buy. Useful business systems are harder to shape. Open-weight releases like Inkling strengthen that argument because they make custom AI more possible, but also make the choices more complex. The opportunity is real, and so is the waste if the work starts in the wrong place.
The next useful AI system will not just be smarter. It will be better fitted to the business it serves.NextAura
The reason this matters now is timing. Open-weight models, hosted fine-tuning platforms and agent tools are all making customisation more accessible at the same moment. That does not make the work simple, but it changes the economics. A system that once needed enterprise budget can now be approached in smaller, focused slices: one sales workflow, one service desk process, one content operation, one internal admin loop.
The businesses that benefit will not be the ones chasing every model release. They will be the ones that know where better fit would pay off: fewer missed leads, faster quoting, cleaner content, stronger AI search visibility, more consistent follow-up, and less time lost moving information between tools. The model is only part of that. The business design around it is where the margin appears.
This is exactly the work NextAura does for Australian small businesses. We build AI agents and automation around the way a business actually runs, choosing the right tools, shaping the workflow and keeping the implementation grounded in commercial outcomes. If open-weight AI has you wondering what should be custom in your business and what should stay simple, get in touch and we will handle the optimising and automating while you stay focused on customers.
Tell us where you are headed. We will come back with a scope, a price, and a launch date you can plan around.
Book a free consultation
StrategyOpenAI's latest guidance says AI value is no longer measured by cheap tokens alone. For small businesses, the real question is who turns AI spend into repeatable work that earns its keep.
StrategyEvery time your team corrects an AI, it reveals judgement competitors cannot buy. A new argument from Satya Nadella reframes those traces as business knowledge worth protecting and compounding.
StrategyGoogle quietly expanded the product data it reads from your website, including a new category field and proper sale start and end dates. For online stores, it is a cheap way to win more visibility in search.