Thinking Machines Ships Inkling, a 975B Open-Weight Model From Mira Murati's Lab
Thinking Machines Lab released Inkling on July 15, 2026 -- its first in-house AI model and the company's clearest statement yet about what it's actually building.
The Model
Inkling is an open-weight mixture-of-experts model. Total parameter count: 975 billion. Active parameters per forward pass: approximately 41 billion. That ratio is the point -- large capacity, lower inference cost.
Training data ran to 45 trillion tokens across text, image, audio, and video. The model reasons natively across all four modalities. Outputs are currently limited to text, code, styled artifacts, and structured data.
On coding benchmarks, Inkling claims roughly one-third the token usage of Nvidia's Nemotron 3 Ultra to reach equivalent performance. That's a meaningful efficiency claim if it holds in real workloads.
How It Was Built
Thinking Machines pre-trained Inkling from scratch on Nvidia GB300 NVL72 systems. For some early post-training data, before scaling up reinforcement learning, they used Moonshot AI's Kimi K2.5 to generate examples. Other open-weight models contributed to post-training data as well.
The company struck a partnership with Nvidia in March to deploy a gigawatt of Vera Rubin computing capacity. Inkling is the first public output of that arrangement.
The Customization Angle
Thinking Machines' model customization platform is called Tinker. The Bridgewater Associates partnership illustrates the pitch: a jointly fine-tuned model on financial expertise reached 84.7% on financial reasoning benchmarks and cost roughly one-fourteenth as much to run as top proprietary alternatives.
The company released a research preview of "interaction models" in May 2026, focused on real-time listen-and-speak AI. Inkling and that work together suggest a strategy aimed at vertical deployment rather than general-purpose competition with frontier labs.
Context
Thinking Machines was founded by Mira Murati, former CTO at OpenAI. The company now employs around 200 people. A reported $50 billion fundraising round was said to be coming together in November 2025 but had stalled by January 2026. Inkling arriving in July suggests the company kept building regardless.
Open-weight release means the model is available for inspection and fine-tuning. Whether the efficiency claims survive contact with production workloads is the next question.
Source: Techcrunch