Thinking Machines Lab Launches Inkling and Inkling-Small Open-Weight Models

Thinking Machines Lab, the AI research startup founded by Mira Murati, former CTO of OpenAI, has launched Inkling, an open-weights AI model built for customizable, domain-specific workflows. Days later, the company followed up with Inkling-Small, a leaner variant designed to bring most of Inkling’s capability to a much smaller footprint.

Tinker: the platform behind the models

Both models ship through Tinker, Thinking Machines Lab’s training and inference platform. Tinker is designed to simplify and automate lightweight fine-tuning — using LoRA adapters — of large open models, giving developers a hosted API for distributed fine-tuning without hiding the underlying knobs.

Inkling-Small: near-Inkling performance at a quarter of the size

Inkling-Small achieves performance near the level of the full Inkling model at just over a quarter of the size:

• 276B total parameters, 12B active — a mixture-of-experts design
• Natively handles text, images, and audio, just like Inkling
• Ships with efficient variable thinking effort, letting the effort level scale with task difficulty
• Available today on Tinker at 64K and 256K context lengths

According to Thinking Machines Lab, Inkling-Small’s efficiency and size make it a cost-effective choice for many tasks, and the company is offering it — alongside Inkling — at a 50% discount for a limited time. Teams looking to get started can follow the Inkling cookbook published alongside the release.

Why it matters

Inkling-Small’s launch, roughly two weeks after Inkling itself, points to a fast iteration cadence: the model reportedly benefited from an improved pre-training data mix and on-policy distillation using Inkling as a teacher, followed by further agentic-coding reinforcement learning — enough to surpass Inkling on some reasoning and agentic-coding benchmarks despite the smaller size. For architects and engineers evaluating open-weight models for fine-tuning, Inkling-Small is a notable option to track alongside other open multimodal releases.

Disclaimer:

All data and information provided on this blog are for informational purposes only. All the image sources used are for reference only. The author makes no representations as to the accuracy, completeness, correctness, suitability, or validity of any information on this blog and will not be liable for any errors, omissions, or delays in this information or any losses, injuries, or damages arising from its display or use. This is a personal view and the opinions expressed here represent my own and not those of my employer or any other organization.