Thinking Machines publishes Inkling 2 Small weights, but local access remains limited

Inkling 2 Small weights are reported available on Hugging Face under Apache 2.0 after a hosted-only debut. The 594B-total/17B-active model’s reported 500–800GB aggregated-VRAM requirements sharply limit practical self-hosting; licensing, capabilities, and external evaluations should be checked against the actual artifacts.