Thinking Machines

Thinking Machines Lab announced Inkling on July 15, 2026 as its first fully open-weights model, a Mixture-of-Experts transformer with 975B total parameters and 41B active, supporting a 1M-token context window and pretrained on 45 trillion tokens of text, images, audio, and video. Alongside the flagship, the lab release Inkling is available for fine-tuning on Tinker starting on launch day, and the company introduced an Inkling Playground inside the Tinker console for chatting with the model. To demonstrate self-customization, Thinking Machines published a session where Inkling wrote and ran its own fine-tuning job on Tinker and evalua

Thinking Machines Lab has released Inkling, a 975-billion-parameter open-weights AI model distributed under the Apache 2.0 license, making it the largest U.S.-origin open-weights model and an explicit alternative to Chinese open models such as DeepSeek V4, GLM 5.2, and Kimi K2.6. The model uses a mixture-of-experts arc Inkling is a chain-of-thought reasoning model trained with reinforcement learning, and Thinking Machines claims it matches Nvidia's Nemotron 3 Ultra on Terminal Bench 2.1 while using about one-third as many thinking tokens — though these figures are vendor-reported and not independently verified in the supplied evidenc

Thinking Machines Lab released Inkling, its first in-house model, as an open-weight Mixture-of-Experts transformer with 975B total parameters, 41B active, supporting up to a 1M-token context window and pretrained on 45 trillion tokens spanning text, images, audio, and video. The company framed Inkling as the first of a Alongside the flagship release, Thinking Machines also previewed Inkling-Small, a lighter-weight 12B-active-parameter model trained with the same recipe to deliver strong performance at lower cost and latency. The launch is positioned within the lab's broader argument that AI trained by a single provider and hardcoded

Thinking Machines Lab released Inkling on July 16, 2026 as its first general-purpose open-weight model, built from scratch under Mira Murati's leadership. The system is a 975-billion-parameter model with 41 billion parameters active per token during inference, supports native text, image, and audio input with text-only Inkling ships with a one-million-token context window and is distributed under an Apache 2.0 license per the published model card, with full weights available for download. Developers can fine-tune the model through Thinking Machines' Tinker platform or deploy it through third-party providers, though GadgetsNow flags t

Thinking Machines Lab, Mira Murati's company, launched Inkling on July 15, 2026 as its first foundation model trained fully from scratch, and the lab is releasing Inkling's full open weights to developers so they can download and fine-tune the model themselves rather than access it only through a hosted API. The releas For developers, the open-weights release means Inkling can be self-hosted, inspected, and adapted to downstream tasks, with fine-tuning supported through Thinking Machines' existing tooling. The SiliconANGLE report frames the launch as a significant open-weights milestone for the lab, though specific technical details

The Wall Street Journal reported July 15, 2026 on Thinking Machines Lab's first model release, confirming Inkling is a 975-billion-parameter open-weights model aimed at chipping away at the lead of frontier labs like OpenAI and Anthropic. The piece frames the launch as Murati's bet on customizable AI over closed "walle WSJ situates the release within a broader industry debate, citing Palantir CEO Alex Karp and Microsoft's Satya Nadella among leaders warning that closed-frontier concentration could undermine the AI ecosystem. For developers and enterprise buyers, the corroboration of an open-weights release from a well-funded Murati-l

Thinking Machines Lab unveiled a research preview of what it calls "interaction models" on May 11, 2026, a new class of native multimodal systems designed to move AI beyond turn-based chat. Unlike today's frontier models that freeze perception while generating a response, these systems treat interactivity as a first-cl The preview is not yet available to the general public or enterprises — Thinking Machines said in its announcement that it will open a "limited research preview" in the coming months, with a wider release planned for later in 2026. This sits as a distinct technical thread from the July Inkling model release and is rele