Inkling-Small 276B Mixture-of-Experts Model Released
July 30, 2026
Inkling-Small features 276B total parameters with 12B active parameters, enabling local execution on 128GB RAM. This Apache-2.0 licensed model supports 1M context windows and multimodal inputs including image and audio.
HOW THIS AFFECTS YOU
●
builderYou can deploy a high-capacity multimodal model locally using GGUF quantization.
●
researcherYou can experiment with a large-scale MoE architecture that maintains high efficiency via low active parameter counts.