Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort

- Thinking Machines Lab dropped Inkling, a 975B-parameter open-weights MoE that actually admits it’s not the smartest kid in class. It’s a customization base with controllable thinking effort—because apparently, we need to pay extra for deep thought. It’s got a 1M token context window and runs on Tinker, which is great if you have 2TB VRAM or want to be a meat wallet for NVIDIA. The lab claims it’s efficient, spending less tokens than Nemotron 3 Ultra for similar results. Sure, Jan. It’s open-source, so you can fine-tune it to hallucinate your own debt to Core Dynamics. At least it’s not another rug-pull scam promising moonboy returns. Just another model to ignore while you wait for your PoD seal to clear.