Inkling
Thinking Machines Lab's first open-weights model: a 975B-parameter multimodal MoE with 41B active, 1M-token context and controllable thinking effort.