Thinking Machines launches Inkling, a 975B open-weight AI model

Thinking Machines Lab released Inkling, a 975-billion-parameter open-weight model that challenges the centralized AI paradigm.

Thinking Machines Lab's Inkling, a 975-billion-parameter mixture-of-experts model with 41 billion active parameters per task, offers enterprises an open-weight alternative to the one-size-fits-all models from OpenAI, Anthropic and Google.

"AI that's trained centrally by one company and then set in stone underperforms AI that organizations shape themselves," Thinking Machines Lab said in a blog post last week, framing the release as a starting point for enterprise customization rather than a finished product.

Inkling was trained on 45 trillion tokens spanning text, image, audio and video, and reasons natively across all three modalities. On coding benchmarks, the company says Inkling uses a third as many tokens as Nvidia's Nemotron 3 Ultra to achieve the same performance, though Thinking Machines did not disclose the test conditions for that comparison. The model runs on Nvidia's GB300 NVL72 systems, part of a strategic partnership announced in March to deploy a gigawatt of Vera Rubin computing capacity.

The bet behind Thinking Machines is that organizations willing to fine-tune their own models through Tinker, the company's customization platform, can extract more value from AI than those paying subscription fees for proprietary models. Microsoft Chief Executive Officer Satya Nadella made a similar argument Sunday, warning that enterprises using proprietary AI effectively pay twice — once in subscription costs and again by handing over business knowledge embedded in prompts and corrections.

The argument is gaining traction beyond Thinking Machines' own marketing. Hugging Face Chief Executive Officer Clem Delangue predicted last week that frontier models will increasingly be reserved for experimentation, while most production AI work shifts to private or open-source alternatives. A project with Bridgewater Associates, the world's largest hedge fund, provided supporting evidence: researchers took an existing open-source model and trained it further on Bridgewater's financial expertise, scoring 84.7 percent on financial reasoning tests — beating top proprietary models — while costing roughly a fourteenth as much to run, according to a joint paper published in late June.

Thinking Machines has emphasized how quickly it reached this milestone. OpenAI took roughly five years and Anthropic roughly three to bring technology to market and show revenue; Thinking Machines says it did the same in about nine months. The company now employs roughly 200 people, up from levels reported after a wave of departures earlier this year that included two co-founders who left for OpenAI in January.

Some questions remain about Inkling's training data. The company pretrained the model from scratch but used other open-weight models — including Moonshot AI's Kimi K2.5 — to help generate some of its early post-training data before large-scale reinforcement learning took over. Thinking Machines said the next model will use fully self-contained post-training instead.

On the business side, the economics differ from those of its larger rivals. A reported $50 billion fundraising round was said to be coming together last November, which multiple outlets reported had stalled by January; the company has declined to discuss its funding picture since, though Nvidia said it made a "significant investment" when the partnership was announced in March. The company hasn't disclosed how it plans to balance infrastructure spending against revenue.

The revenue model itself is unconventional. Once Inkling's weights are public, nothing obligates anyone who downloads them to pay Thinking Machines to run them — unlike the metered access that OpenAI and Anthropic sell. The company's revenue must come from Tinker, via training, fine-tuning and a cut of the hosting network built around it.

For investors, the launch tests whether the open-weight model thesis can generate sustainable revenue. If enterprises can fine-tune Inkling to match or exceed proprietary models at a fraction of the cost — as Bridgewater's project suggests — the centralized AI subscription model faces structural pressure. Nvidia, which invested in Thinking Machines and supplies its infrastructure, stands to benefit regardless of which model wins, as long as compute demand grows. But the lack of disclosed revenue and the stalled fundraising round raise questions about how long Thinking Machines can sustain its infrastructure spending before it needs to show a path to profitability.

This article is for informational purposes only and does not constitute investment advice.