Thinking Machines debuts Inkling, a 975B open multimodal model

Inkling is a 975B-parameter MoE (41B active) open model with native image, text, and audio input, 1M context, trained on 45T tokens. Thinking Machines also released Inkling-Small (276B total, 12B active) with day-0 support in transformers, SGLang, vLLM, and llama.cpp.