Thinking Machines Lab released Inkling, a 975-billion-parameter multimodal model with 41 billion parameters active per token, under Apache-tagged weights with a separate Model Acceptable Use Policy. Despite the open weights, the official BF16 checkpoint occupies about 1.905 terabytes, and the smaller NVFP4 package is 592 gigabytes, requiring clusters of four to sixteen high-end data-center GPUs to run. Inkling is a 66-layer, decoder-only sparse Mixture-of-Experts transformer where each token is routed to six of 256 specialist experts with two shared experts active. The model alternates five local-attention layers with one global-attention layer, uses eight key-value heads, relative positional embeddings, and a maximum context of 1,048,576 tokens. Thinking Machines offers managed LoRA fine-tuning through Tinker at 64K and 256K context. Provider APIs arrived quickly, but several major self-hosted runtime integrations were still open or draft changes during the launch window.