Nex

Nex is an open-source agentic model and infrastructure project, and this launch introduces the Nex-N2.5 family, its next-generation family of agentic models built for long-horizon tasks in real-world environments . The release ships in three sizes aimed at builders of autonomous workflow agents: Nex-N2.5-mini and Nex-N2.5-Pro continue to build on the multimodal foundations of Nex-N2, with focused improvements in computer use, web browsing, and visually grounded agentic capabilities, while Nex-N2.5-Max is built on a 1.6-trillion-parameter, text-only Mixture-of-Experts (MoE) foundation model . For teams evaluating open alternatives to closed frontier models for coding and agent workflows, the trillion-parameter Max is the notable data point, positioned by the team as their first complete post-training run at that scale. The design intent is agents that keep working through multi-step tasks rather than one-shot completions. The models can operate computers and browsers, as well as autonomously execute and test programs , using vision as a live feedback channel to verify outcomes and self-correct. Model weights are being released openly alongside hosted services, with the mini variant published on Hugging Face under Apache-2.0, giving developers a route to fine-tune or self-host without licensing friction. Nex itself is not a single startup but a coalition. It is an innovation alliance initiated by the Shanghai Innovation Institute , released jointly with partners including Shanghai Miracle Plus Intelligence Summit, Moosi Intelligence, Kuafu Technology, and others . The broader stack around the models includes the NexAU agent framework and the NexRL reinforcement learning training framework, positioning this launch less as a standalone model drop and more as an attempt to ship a full open agent platform that founders and research teams can pick up and run.