Mira Murati's Thinking Machines Ships Inkling, a 975B-Parameter Open-Weight Model
Thinking Machines Lab released its first in-house model, an Apache 2.0 mixture-of-experts system with 41B active parameters and a 1M-token context window, positioned as a customizable base rather than a top-benchmark chaser.
Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released Inkling on July 15, its first public model and the company's opening move after 18 months of quiet infrastructure work. It ships as full open weights under Apache 2.0, with a smaller preview variant, Inkling-Small, alongside it.
- 975B total parameters, mixture-of-experts, 41B active (256 routed experts plus 2 shared, 6 active per token)
- 1M-token context window, pretrained on 45 trillion tokens of text, images, audio and video
- 77.6% on SWE-Bench Verified, 97.1% on AIME, 87.2% on GPQA Diamond at effort=0.99
- Inkling-Small preview: 276B total, 12B active parameters
- Roughly 1/14th the inference cost of top proprietary models, per the company
- Available via Tinker, Hugging Face, TogetherAI, Fireworks, Modal, Databricks and Baseten
AI that organizations can adapt for themselves will outperform the one-size-fits-all models the dominant labs currently distribute. - Thinking Machines Lab
The company is explicit that Inkling is not a frontier-beating model: it's pitched as a base for fine-tuning through Tinker, its customization platform, which is where Thinking Machines expects to make money rather than from model licensing. See how it stacks up against other open-weights coding models.