Nex-AGI Releases Nex-N2.6: A 321B Parameter Multimodal Agentic Model
Nex-AGI has released Nex-N2.6, an FP8-quantized 321B parameter multimodal agentic model built on GLM-5.3-Flash. The model is specifically designed for computer use, coding, and long-horizon tool use. This release provides the open-source community with a massive multimodal agent capable of handling complex computer-use workflows and long-horizon tasks. Improvements in reducing repetitive reasoning help bridge the gap toward more reliable autonomous AI agents. The model repository contains the FP8-quantized weights of Nex-N2.6, totaling approximately 321 billion parameters. Following the previous Nex-N2.5-Pro version, this update specifically targets reducing repetitive reasoning and improving adherence to environment tools.
## BACKGROUND
FP8 quantization uses 8-bit floating-point formats to represent model weights and activations, reducing memory footprint and computational requirements while accelerating inference. Long-horizon agentic models are designed to execute extended, interdependent action sequences and maintain goals over multi-step interactions.