Thinking Machines Lab Launches Inkling 2 Small with Open Weights Pending

Thinking Machines Lab launched Inkling 2 Small, a 594B-parameter sparse Mixture-of-Experts model with a 1M-token context window, available now on Tinker with closed weights and planned open weights after safety review. The model supports text and images, multitasks including coding and vision, and was trained extensively on NVIDIA hardware with reinforcement learning. Open weights and fine-tuned checkpoints will be released under Apache 2.0 following external safety testing and red-teaming.