Thinking Machines Launches Inkling 1T Multimodal Model with Unique Architecture

Thinking Machines released Inkling, a new 1T multimodal model capable of reasoning across text, image, and audio, making its full weights available for fine-tuning today. Analysis of Inkling's J-space geometry shows it maintains a consistent structure across its stack, unlike other models that split into distinct sensory, workspace, and motor blocks. Further analysis on Poolside AI's Laguna XS 2.1 model demonstrated that quantization (bf16 vs nvfp4) had almost no impact on its J-lens space.