Mira Murati's AI Startup Releases First Model in Bid to Loosen AI Giants' Grip
The Wall Street Journal ● Covered by 3 sources
Mira Murati's Thinking Machines Lab just dropped its first model, called Inkling. It's a 975-billion-parameter generalist built to compete with the big labs' foundation models.
Based on reporting by The Wall Street Journal — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Mira Murati left OpenAI with a lot of fanfare and even more funding, and now Thinking Machines Lab has something to show for it. The company's first model, Inkling, weighs in at 975 billion parameters, which puts it firmly in the same weight class as the frontier models from OpenAI, Google, and Anthropic. This isn't a narrow tool built for one task. It's pitched as a broad, balanced foundation model meant to hold its own across coding, reasoning, and general knowledge work without falling apart the moment you push it outside its comfort zone.
What's notable is the framing. Thinking Machines isn't chasing a single benchmark headline. Instead, the pitch is about balance, specifically the tradeoff between cost and capability. That's a real gap in the market right now. A lot of enterprises don't need the absolute smartest model on the planet, they need something that's good enough, cheap enough, and controllable enough to run at scale. Inkling seems aimed squarely at that buyer.
The more interesting piece, though, might be Tinker, the cloud-based fine-tuning layer that ships alongside the model. Raw foundation models are rarely what companies actually deploy. They get adapted, narrowed, and reshaped for specific jobs. Tinker is Thinking Machines' bet that owning the customization layer, not just the base model, is where the real value and the real lock-in lives. Every major lab is converging on this idea, but few have made it this central to a first release.
Murati has assembled a team stacked with OpenAI and Google DeepMind alumni, and the expectations attached to this launch reflect that pedigree. A 975-billion-parameter model with its own fine-tuning ecosystem is a serious first move, not a toy demo. Whether it actually loosens the grip of the incumbents depends on adoption, pricing, and whether Tinker becomes a genuine developer habit rather than a nice-to-have feature.
My take — AI-written commentary, not fact-checked reporting
I'll believe the 'loosening the giants' grip' narrative when I see pricing and benchmarks that aren't cherry-picked by a comms team. Ex-OpenAI pedigree buys attention, not market share, and the AI space is littered with well-funded generalist models that never found a home. Tinker is the smarter bet here, honestly, because owning the fine-tuning layer is a stickier business than shipping yet another giant foundation model.
Read more about this at: The Wall Street Journal