23 Downloads Updated 2 months ago
Nova AI is a powerful all in one AI model that is excellent for performance while maintaining less Requirements.
ollama run relational/nova:7b
Applications
ollama launch claude --model relational/nova:7b
ollama launch codex-app --model relational/nova:7b
ollama launch openclaw --model relational/nova:7b
ollama launch hermes --model relational/nova:7b
ollama launch codex --model relational/nova:7b
ollama launch opencode --model relational/nova:7b
Models
View all →Readme
Nova Helix
Advanced Coding & Infrastructure AI Models
Built for practical engineering, production workflows, and modern development.
Models
| Model | Parameters | Focus | Runtime |
|---|---|---|---|
| Nova Helix Coder 7B | 7.7B | Lightweight coding + infrastructure | Ollama |
| Nova Helix 2.5 | 31.2B | Advanced reasoning + engineering | Ollama |
Nova Helix Coder 7B
Lightweight. Fast. Practical.
Nova Helix Coder 7B is built on top of Qwen2.5-Coder-7B-Instruct and optimized for real-world software engineering.
Features
- Modern website generation
- React / Tailwind / TypeScript
- Backend APIs
- Docker & Linux
- nginx & infrastructure
- Roblox Lua scripting
- Debugging & optimization
- Ollama infrastructure
- Production-ready code generation
Model Information
| Property | Value |
|---|---|
| Base Model | Qwen2.5-Coder-7B-Instruct |
| Fine-Tune Method | QLoRA |
| Total Parameters | 7.7B |
| Trained Parameters | ~80.7M |
| Context Length | 8192 |
| Quantization | Q4_K_M |
| Runtime | Ollama |
Nova Helix 2.5 (32B)
Advanced reasoning. Elite engineering.
Nova Helix 2.5 is a next-generation coding and infrastructure AI based on GLM-4.7-Flash and optimized for practical engineering workflows.
Nova 2.5 focuses on:
- advanced coding
- debugging
- frontend generation
- backend systems
- infrastructure reasoning
- Docker/Linux/nginx
- AI hosting
- DevOps workflows
- production engineering
Nova Helix 2.5 Features
- Advanced software engineering
- React / Next.js / Tailwind
- Python / Node.js / Go / Rust
- Backend APIs & databases
- Docker / Linux / nginx / Redis
- Infrastructure debugging
- Long-context engineering workflows
- Better frontend/UI generation
- Practical architecture reasoning
- Production-ready outputs
Nova Helix 2.5 Model Information
| Property | Value |
|---|---|
| Runtime Name | nova:32b |
| Base Model | unsloth/GLM-4.7-Flash |
| Architecture | GLM4 MoE Lite |
| Fine-Tune Method | QLoRA |
| Total Parameters | ~31.2B |
| Estimated Trained Parameters | ~160M–320M |
| Context Length | 8192 |
| Quantization | Q4_K_M |
| Runtime | Ollama |
| Recommended GPU | RTX 4090 / RTX A6000 |
Estimated Engineering Performance
| Task | Rating |
|---|---|
| Website Generation | 9.4⁄10 |
| Frontend UI/UX | 9.5⁄10 |
| React + Tailwind | 9.5⁄10 |
| Docker / Linux | 9.3⁄10 |
| nginx / Reverse Proxy | 9.3⁄10 |
| Backend APIs | 9.2⁄10 |
| Infrastructure | 9.3⁄10 |
| Debugging | 9.1⁄10 |
| Architecture Planning | 9.0/10 |
| Technical Reasoning | 9.1⁄10 |
| General Coding | 9.2⁄10 |
Benchmark Comparison
NovaBench v2 — Practical Engineering Benchmark
| Model | Web Dev | Infra | Debugging | Reasoning | UI Quality | Speed |
|---|---|---|---|---|---|---|
| Nova Helix 2.5 (32B) | 9.4 | 9.3 | 9.1 | 9.1 | 9.5 | 8.4 |
| Claude Opus 4.1 | 9.8 | 9.7 | 9.8 | 9.9 | 9.6 | 7.8 |
| GPT-4.1 | 9.6 | 9.4 | 9.5 | 9.7 | 9.2 | 8.3 |
| Gemini 2.5 Pro | 9.5 | 9.2 | 9.3 | 9.7 | 9.1 | 8.6 |
| DeepSeek R1 | 9.1 | 8.9 | 9.2 | 9.5 | 8.5 | 7.2 |
| Nova Helix Coder 7B | 9.1 | 8.9 | 8.5 | 8.2 | 9.3 | 9.0 |
| Qwen2.5-Coder-7B | 8.8 | 8.1 | 8.3 | 8.0 | 8.4 | 9.1 |
| DeepSeek Coder 6.7B | 8.5 | 7.9 | 8.1 | 7.8 | 7.8 | 8.7 |
| CodeLlama 7B | 7.8 | 7.2 | 7.5 | 7.1 | 7.0 | 8.9 |
| Mistral 7B Instruct | 7.6 | 7.1 | 7.3 | 7.3 | 7.2 | 9.2 |
Estimated Technical Benchmarks
| Benchmark | Nova Helix 2.5 |
|---|---|
| HumanEval | 86–92% |
| MBPP | 82–89% |
| MultiPL-E | 78–86% |
| SWE-style Debugging | 55–70% |
| Instruction Following | 9.2⁄10 |
| Infrastructure Tasks | 9.3⁄10 |
| Frontend Generation | 9.5⁄10 |
Example Prompts
Build a modern SaaS dashboard with authentication, PostgreSQL, and responsive design using Next.js and Tailwind.
Fix this nginx reverse proxy for Ollama behind Cloudflare and optimize security.
Create a production Docker Compose stack with Redis, MariaDB, nginx, and a Node.js API.
Analyze this Linux server log and explain exactly what is failing.
Optimize this Ollama deployment for an RTX A6000 with multiple models.
Ollama
ollama run nova:7b-q4
ollama run nova:32b
Hardware Recommendations
| Model | RAM Needed | VRAM Needed | Recommended Hardware |
|---|---|---|---|
| Nova 7B Q4_K_M | 8GB–12GB | 6GB+ | RTX 2060 / Arc A380 |
| Nova 32B Q4_K_M | 24GB–40GB | 16GB–24GB+ | RTX 4090 / A6000 |
| Nova 32B Q5_K_M | 32GB–48GB | 24GB+ | A6000 / multi-GPU |
| Nova 32B Q8_0 | 64GB–96GB | 32GB+ | High-end servers |
| Nova 32B F16 | 96GB+ | 48GB+ | Enterprise/server use |
Philosophy
Built for real engineering.
Nova Helix focuses on practical development instead of benchmark-only optimization.
Designed for:
- modern web development
- infrastructure engineering
- backend systems
- AI hosting
- debugging
- deployment
- production workflows
Roadmap
- Nova Helix 2.5 GGUF
- Better SWE-style training
- Expanded frontend generation
- Better autonomous debugging
- Long-context engineering workflows
- Vision-language support
- Multi-agent infrastructure tools
- Nova flagship 120B experiments
License
Open-source release by Orbis.
Community improvements, datasets, and fine-tunes are welcome.