Nex-N2

Nex-AGI's open-weight agentic model family, post-trained on the Qwen3.5 MoE series for first-tier coding and long-horizon agentic work. Two Apache-2.0 variants: Nex-N2-Pro (built on Qwen3.5-397B-A17B — 397B total / 17B active) and Nex-N2-mini (built on Qwen3.5-35B-A3B-Base — 35B / 3B active), both multimodal (image/video input).

Its core is an "Agentic Thinking" framework that unifies reasoning, tool use, and environment execution into one closed loop: Adaptive Thinking (the model decides when and how deeply to think) and Coherent Thinking (one reasoning paradigm transferred across general and agentic tasks). Trained via Nex-AGI's own NexRL stack and the environment-scaling recipe from Nex-N1. Self-reported Nex-N2-Pro: Terminal-Bench 2.1 75.3, SWE-Bench Verified 80.8, SWE-Bench Pro 58.8, GDPval 1585, TAU3 71.1 — pacing GPT-5.5 / Opus 4.7 on coding and long-horizon tasks. Not yet on the AA Intelligence Index.