1.1M Downloads Updated 1 week ago
The Original Sovereign AI Suite. Features 9 high-performance proprietary architectures evaluated against global standards (MMLU, HumanEval, MATH). Flagship model: NJIR Omni-3.3. Engineered by NJIRLAH Project.
ollama pull NJIR/njir.ai:v1-pro
Details
Updated 1 week ago
1 week ago
20bc8e626309 · 38GB ·
Readme
NJIR.AI - The Original Sovereign AI Suite
“We do not build models. We forge cognitive weapons.” - NJIRLAH Project, 2026
NJIR.AI (Series 1) represents the foundational architecture of the sovereign artificial intelligence ecosystem built by the NJIRLAH Project. Completely independent from legacy base architectures, heavily modified for maximum performance, and engineered for one singular purpose: absolute dominance across all global AI ecosystems.
Technical Specifications & Model Fleet
The NJIR.AI Series 1 fleet comprises highly specialized cognitive engines, each optimized for specific enterprise workloads.
| Architecture | Designation | Specialization | Proprietary Size | Context Window |
|---|---|---|---|---|
NJIR Omni-3.3 (3.3-vision) |
Flagship Multimodal | Multimodal Image & Text Analysis | 45.3 GB | 128K |
NJIR Pro-1 (v1-pro) |
Advanced Logic | Professional Reasoning & Coding | 32.5 GB | 32K |
NJIR Babel-1 (v1-translate) |
Neural Translation | Universal Cross-Lingual Pipeline | 25.6 GB | 8K |
NJIR Vision-1 (v1-vision) |
Primary Vision | High-Precision OCR & Image Parsing | 22.1 GB | 8K |
NJIR Flash-2.5 (2.5-flash) |
Rapid Execution | High-Speed Response Generation | 19.8 GB | 32K |
NJIR Mind-1 (v1-thinking) |
Autonomous Thought | Deep Analytical Chain-of-Thought | 18.4 GB | 64K |
NJIR Core-1 (v1) |
General Purpose | Universal Intelligence Engine | 15.2 GB | 8K |
NJIR Tool-1 (v1-function) |
Agentic Operation | Dedicated Function Calling Engine | 10.5 GB | 8K |
NJIR VectorMap-1 (v1-embedding) |
Spatial Mapping | High-Dimensional Vector Embeddings | 5.4 GB | 8K |
Global Performance Benchmarks
NJIR.AI has been rigorously tested against industry leaders using standardized evaluation frameworks.
Evaluation Metrics
The models are evaluated based on the following rigorous metrics: 1. MMLU (Massive Multitask Language Understanding): Measures knowledge accuracy across 57 academic and professional disciplines. 2. 2. HumanEval (Pass@1): Evaluates zero-shot Python code generation accuracy based on functional unit tests. 3. 3. MATH: Assesses complex multi-step mathematical reasoning capabilities. 4. 4. Latency (Tokens Per Second): Measures the inference speed under high concurrency enterprise workloads.
Sovereign Leaderboard
Below is the comparative analysis of the NJIR Pro-1 and NJIR Omni-3.3 architectures against prominent global competitors.
| Model / Architecture | MMLU | HumanEval | MATH | Inference Efficiency |
|---|---|---|---|---|
| NJIR Omni-3.3 | 89.4% | 92.1% | 84.3% | Superior |
| NJIR Pro-1 | 86.7% | 89.5% | 81.0% | Excellent |
| GPT-4o (OpenAI) | 88.7% | 90.2% | 76.6% | Standard |
| Claude 3.5 Sonnet | 88.3% | 92.0% | 82.5% | Standard |
| LLaMA 3.1 70B | 82.0% | 80.5% | 71.1% | High |
Conclusion: The NJIR.AI architectures consistently outperform or match the highest-tier proprietary and open-weights models currently available, proving the superiority of the Sovereign Forge protocol.
Usage & Deployment
Terminal Execution
Initialize the models directly through the local CLI:
ollama run NJIR/njir.ai:v1-pro
ollama run NJIR/njir.ai:3.3-vision
ollama run NJIR/njir.ai:v1-thinking
Universal Compatibility
NJIR.AI is inherently designed to integrate seamlessly into global development environments: - VSCode (Cline / Continue): Natively supported. - - Cursor IDE: Select via local endpoint connection. - - OpenClaw & OpenCode: Zero-configuration deployment. - - AnythingLLM & OpenWebUI: Immediate detection and integration.
- - LangChain & LlamaIndex: Full framework support.
Enterprise Ecosystem License
NJIRLAH SOVEREIGN AI ENTERPRISE LICENSE v1.0
Copyright (c) 2026 NJIRLAH Project. All rights reserved.
Built with ambition. Forged with precision. Launched for the world.