Inkling 2 Small launches on Tinker for inference and fine-tuning, with 594B total/17B active parameters and image input. Thinking Machines plans Apache 2.0 weights next week, pending external safety tests and red-teaming.