eadx/Nex-N2-mini

TEXT GENERATIONConcurrency Cost:3Model Size:35.1BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jun 9, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

Nex-N2-mini is a 35.1 billion parameter agentic language model developed by Nex-AGI, built upon the Qwen3.5-35B-A3B-Base architecture. It is specifically designed for real-world productivity scenarios, excelling in agentic tasks, coding, and long-horizon problem-solving through its Agentic Thinking framework. This model unifies reasoning, tool use, and environment execution to deliver stable, end-to-end results in complex tasks.

Loading preview...

Nex-N2-mini: An Agentic Model for Real-World Productivity

Nex-N2-mini, developed by Nex-AGI, is a 35.1 billion parameter agentic language model based on the Qwen3.5-35B-A3B-Base series. It is engineered to tackle complex, long-horizon tasks by integrating reasoning, tool use, and environmental execution into a unified "Agentic Thinking" framework. This framework comprises Adaptive Thinking, allowing the model to dynamically adjust its reasoning depth, and Coherent Thinking, which maintains a consistent reasoning paradigm across diverse tasks and modalities.

Key Capabilities

  • Agentic Workflows: Excels in tasks requiring environmental feedback, such as agentic coding, deep research, tool calling, and terminal execution.
  • Coding & Software Engineering: Demonstrates strong performance on benchmarks like Terminal-Bench 2.1 (60.7) and SWE-Bench Pro (50.2), indicating robust capabilities in software development tasks.
  • General Reasoning: Maintains competitive performance in general capability and core reasoning tasks, standing on par with leading frontier models in its class.
  • Function Calling & Reasoning Traces: Supports robust function-calling and emits explicit reasoning traces, enhancing transparency and control over its decision-making process.

Good for

  • Automating Complex Tasks: Ideal for scenarios requiring models to execute multi-step processes with environmental interaction and feedback.
  • Software Development: Suitable for developers needing assistance with coding, debugging, and managing software engineering workflows.
  • Research & Decision-Making: Can be applied to tasks demanding deep research and search-based decision-making, leveraging its agentic capabilities.
  • Resource-Efficient Agentic Applications: As the 'mini' variant, it offers a balance between performance and computational efficiency for agentic workloads compared to larger models.

Finetunes
1 models