Inkling

Inkling is designed as a general-purpose multimodal autoregressive transformer for applications that combine text with images or audio and produce text responses. Its documented architecture is a 66-layer decoder-only model, and the available model-card and catalog descriptions support its use across English, other languages, and multiple programming languages. Distributed builds include BF16 and NVFP4 variants, with the NVFP4 artifacts shown as signed in the catalog, giving developers options for different deployment requirements.

The model is intended for developers building coding assistants, agentic and tool-use systems, chatbots, retrieval-augmented generation systems, instruction-following applications, and general conversational or multimodal experiences. That combination makes it a practical fit when one model must handle varied inputs and text-based interaction rather than being limited to a single task. The available evidence does not provide supported benchmark results or verified training-scale figures, so comparisons should focus on its documented use cases and deployment format.