aiqbench takeAI Products

Thinking Machines Lab's release of Inkling-Slow demonstrates a significant shift in accessibility for large-scale Mixture-of-Experts models by enabling inference on consumer-grade hardware like the NVIDIA B300 via specialized NVFP4 quantization. This architecture allows startups and regulated industries to self-host 276B parameter multimodal systems without relying solely on frontier data center clusters, effectively lowering the barrier to entry for advanced reasoning tasks across text, audio, and vision.

Source coveringMarkTechPost

What this isA model's reading of the article and related coverage, written for aiqbench. Not the publisher's words.

Thinking Machines Lab Releases Inkling-Small: A 276B Total, 12B Active Open Weights Multimodal MoE Model

MarkTechPostAugust 2, 20264 min read

Inkling-Small matches Inkling at a quarter the size, and its NVFP4 checkpoint runs on one NVIDIA B300 GPU The post Thinking Machines Lab Releases Inkling-Small: A 276B Total, 12B Active Open Weights Multimodal MoE Model appeared first on MarkTechPost.

Thinking Machines Lab has released Inkling-Small, an open weights Mixture-of-Experts model with 276B total parameters and 12B active. That is about a quarter the size of Inkling, which carries 975B total and 41B active parameters. The model was trained on NVIDIA GB300 NVL72 systems. It reasons natively over text, images and audio. The context window reaches 1M tokens, and thinking effort is adjustable. Weights ship under Apache 2.0 on Hugging Face. Yes, and the quantized checkpoint is why. Per the model card,…

Excerpt shown under the publisher's RSS terms · 82 words · full text and figures are on the publisher's site

Read at MarkTechPost ↗Ask about this article