Descriptions:
Fahd Mirza reviews Nex-N2 Pro, the second-generation agentic model from Next AGI — an open-source collaboration between Shanghai Innovation Institute, Fudan University, and several Shanghai-based AI companies. Built on a nearly 400-billion-parameter Qwen 3.5 mixture-of-experts base, the model is explicitly designed for real-world agentic tasks rather than general-purpose chat. A lighter Nex-N2 Mini variant is also available for latency-sensitive deployments. Both models are fully open-sourced and currently free to access via hosted sources including OpenRouter.
The video’s centerpiece is a live agentic test using the Hermes coding agent: the model is tasked with building a full-stack banking dashboard in Django REST Framework and React, with five intentional bugs planted across dependency management, backend config, database state, auth middleware, and frontend configuration. Over roughly 36 minutes and 66,000+ tokens, Nex-N2 Pro autonomously wrote, debugged, and tested the full codebase — including generating its own test suites and iterating on failures.
Benchmark analysis shows the model leads on terminal execution tasks and performs respectably across coding and reasoning suites, while trailing frontier proprietary models like GPT and Claude Opus on the most demanding software engineering evaluations. A standout visualization tracks the model’s iterative progress writing a SQL interpreter from scratch, illustrating self-improvement across multiple debugging loops — something static benchmarks cannot capture. Mirza also flags that the role of the software engineer is shifting toward supervision and redirection rather than direct code authorship.
📺 Source: Fahd Mirza · Published June 11, 2026
🏷️ Format: Review
