[REVISION HISTORY]

Thinking Machines Lab Inkling AI launch

Updated 1 time since CLSTR started tracking revisions of this situation.

What changed

2026-07-27 05:03 UTC → 2026-07-31 19:09 UTC · added ~~removed~~

In mid‑July 2026, Thinking Machines Lab, the startup founded by former OpenAI CTO Mira Murati, released Inkling, a 975‑billion‑parameter multimodal mixture‑of‑experts model that activates roughly 41 billion parameters per query and supports a one‑million‑token context window. The launch was officially announced on 15 July 2026, and the model’s weights were made available under an Apache 2.0 licence on Hugging Face. The companion Tinker platform enables enterprises to fine‑tune Inkling on proprietary data for customized behaviour. Strategic infrastructure partners underpin the effort: Nvidia supplies gigawatt‑scale Vera Rubin compute capacity under a long‑term agreement, while Google Cloud has signed a multibillion‑dollar cloud‑services contract to support training and deployment. These collaborations are intended to position Inkling as an open‑weight, cost‑effective alternative to large proprietary models. Regional coverage in the Balkans highlighted the launch alongside Vodafone Albania’s pilot of a “Physical AI” system that autonomously repositions 4G and 5G antennas using AI‑driven sensors and robotic arms, illustrating a shift toward adaptive, efficient AI applications. Independent testing reported on 25 July 2026 showed Inkling achieving high benchmark scores (74.1 % on MCP Atlas and 77.6 % on SWE‑Bench) but struggling with complex coding, creative‑writing, and travel‑planning tasks. The model requires enterprise‑grade hardware and is priced at roughly $1 per million input tokens via OpenRouter, emphasizing privacy‑focused, compliance‑oriented use cases rather than everyday developer workloads. Later in July, the lab released Inkling Small, a 276‑billion‑parameter multimodal model with 12 billion active parameters. Weights were posted on Hugging Face on 30 July under an Apache 2.0 licence. The variant scores 40 on the Artificial Analysis Intelligence Index, outperforms the larger Inkling on several coding and reasoning benchmarks, and offers a 256 K‑token context window (noted as 1 million tokens in Japanese release notes).

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.