Welcome Inkling by Thinking Machines

Inkling is a large, open model with approximately 1 trillion parameters and a 1 million token context window that natively accepts image, text, and audio inputs. Developed by Thinking Machines, it was trained on 45 trillion tokens and is designed for multimodal reasoning and domain adaptation. Its architecture is a decoder-only Mixture-of-Experts (MoE) model featuring relative attention, hybrid sliding window attention, and a short 1D convolution for processing. The model is available for inference with day-one support in libraries such as transformers, SGLang, and vLLM.