← All companies

Frontier / Official source checked

Thinking Machines Lab

What Thinking Machines Lab does

Thinking Machines Lab is an AI research and product company focused on (1) frontier-capable models that can be customized for specific users and domains, and (2) human–AI collaboration via “interaction models” that are designed to handle live, multimodal back-and-forth rather than relying on slow, turn-based prompting.

The company’s product strategy centers on a customization and fine-tuning workflow called Tinker, which is designed to let developers and research teams train/fine-tune models while handling distributed training complexity. Thinking Machines’ current open-weights model release is Inkling, a multimodal Mixture-of-Experts transformer (975B total / 41B active) with a long context window and “controllable thinking effort,” meant to serve as a flexible base for further fine-tuning on Tinker.

In parallel, Thinking Machines is building interaction-model research as a long-term bet on scalable human participation: the goal is to make the model itself a native collaborator in real time across modalities, with human knowledge and judgment continuously influencing the system.

Business model (as evidenced by public materials): Thinking Machines emphasizes distribution of model weights and developer-facing tooling for customization (Tinker, including a developer console for interacting with models). It also pursues infrastructure partnerships to support frontier training at scale, including a long-term gigawatt-scale partnership with NVIDIA announced in March 2026.

News

Company record

Jul 30, 2026 · Official · Thinking Machines LabIntroducing Inkling-Small

Thinking Machines released Inkling-Small, an efficient open-weights Mixture-of-Experts model (276B total / 12B active) positioned as achieving comparable performance to Inkling at a quarter of the size, including variable thinking effort and multimodal reasoning; the post also describes evaluation and pricing/cost-per-token framing for output.

Jul 15, 2026 · Official · Thinking Machines LabInkling: Our Open-Weights Model

Thinking Machines released Inkling with full open weights and described the model’s architecture and capabilities, including its multimodal reasoning, long context, controllable thinking effort, and availability for fine-tuning on Tinker; the release also includes Inkling-Small preview material and integration into the Tinker workflow.

Jun 30, 2026 · Official · Thinking Machines LabLearning to Replicate Expert Judgment in Financial Tasks

The company published a research item describing methods for learning to replicate expert judgment in financial tasks, positioned as part of its broader effort around customizable systems and judgment/epistemics.

May 19, 2026 · Official · Thinking Machines LabAnnouncing Interactivity Research Grants

Thinking Machines announced interactivity-focused research grants, encouraging proposals aligned with its “interaction models” direction for real-time human–AI collaboration.

Mar 19, 2026 · Official · Thinking Machines LabTraining LLMs to Predict World Events (Guest Post with Mantic)

Thinking Machines published a guest-style post about training LLMs to predict world events, in collaboration with Mantic, as part of its ongoing emphasis on judgment and forecasting-style capabilities.

Mar 10, 2026 · Official · Thinking Machines LabThinking Machines Lab and NVIDIA Announce Long-Term Gigawatt-Scale Strategic Partnership

The company announced a multi-year strategic partnership with NVIDIA to deploy at least one gigawatt of next-generation NVIDIA Vera Rubin systems to support Thinking Machines’ frontier training and customizable AI platforms; the company also stated NVIDIA made a significant investment to support its long-term growth.

Dec 12, 2025 · Official · Thinking Machines LabTinker: General Availability and Vision Input

Thinking Machines announced General Availability for Tinker and described added support for vision input, furthering the platform’s goal of making customization workflows accessible for developers and researchers.

Nov 07, 2025 · Official · Thinking Machines LabTinker: Call for Community Projects

Thinking Machines issued a community call for projects related to Tinker, aiming to support external experimentation using its customization platform.

Show 2 earlier updates
Source map · 5 recurring channels · 12 references

What it builds

Tinker

Tinker is Thinking Machines’ customization and fine-tuning platform that lets users run and manage model training/fine-tuning workflows, including using the company’s open-weights models as bases; it also includes a console experience for interacting with models during development and evaluation.

source ↗
Inkling

Inkling is Thinking Machines’ open-weights frontier model (975B total / 41B active) designed for multimodal reasoning over text, images, and audio with long context and controllable “thinking effort,” positioned as a customizable base model available for fine-tuning on Tinker.

source ↗
Inkling-Small

Inkling-Small is Thinking Machines’ companion open-weights model (276B total / 12B active) released as an efficient variant positioned as achieving comparable performance to Inkling at a quarter of the size, with variable thinking effort and multimodal reasoning.

source ↗
Interaction models

Interaction models are Thinking Machines’ research direction for AI systems trained to handle live, multimodal human–AI interaction natively (as opposed to being orchestrated externally via turn-based scaffolding).

source ↗

Milestones & partnerships