I have been reading about Inkling, the new open-weight model from Thinking Machines (from ex-OAI Mira Murati), and I am curious whether anyone here has spent real time chatting with it. The specs sound interesting. It is multimodal, has 975B total parameters with 41B active, and supports a context window of up to 1M tokens. But specs and benchmarks never really tell you what a model feels like in normal use. How is it for writing, coding, reasoning, and ordinary conversation? Does it have a ...