Nex-N1
Nex-N1 is Nex AGI’s first published family of agentic language models, described in its December 4 technical report. Post-training emphasizes tool use, coding, research, and sustained task execution across generated environments. The family spans InternLM3-8B, Qwen3-30B-A3B, Qwen3-32B, and a DeepSeek-V3.1 flagship; the different backbones and sizes are grouped here. Apache-licensed model cards, weights, and training data are available. The later DeepSeek-V3.1-Nex-N1.1 checkpoint extends context to 200K and is recorded as a significant family variant. Model-generated actions require separate agent tooling and environmental feedback.
2025-12-04
8B to 671B
Dense and Mixture-of-Experts decoder-only Transformers
Apache-2.0
Specifications
- Parameters
- 8B to 671B
- Architecture
- Dense and Mixture-of-Experts decoder-only Transformers
- License
- Apache-2.0
- Type
- text
- Modalities
- text
Benchmark Scores
Advanced Specifications
- Model Family
- Nex
- API Access
- Not Available
- Chat Interface
- Not Available
- Variants
- internlm3-8B-Nex-N1Qwen3-30B-A3B-Nex-N1Qwen3-32B-Nex-N1DeepSeek-V3.1-Nex-N1DeepSeek-V3.1-Nex-N1.1 (200K context)
Capabilities & Limitations
- Capabilities
- agentic reasoningcodingtool useresearch
- Known Limitations
- Parameter and context limits differ by backboneLarge flagship weights require distributed servingAgent benchmarks depend on tools, scaffolds, and execution budgets
- Notable Use Cases
- local research agentsrepository coding assistanceagent-training research
- Tool Use Support
- Yes