🔱 SKT OMNI SUPREME: The 1.1-Trillion Parameter Frontier
SKT AI LABS
The Sovereign LLM Development For India (Project Surya)
-orange?style=for-the-badge&logo=target){kind=icon}
🌌 Introduction
SKT OMNI SUPREME is the flagship achievement of Project Surya. Developed by Shrijan Kumar Tiwari and the SKT AI Labs team, this is a ground-up pre-trained (Scratch) 1.1-Trillion parameter Mixture-of-Experts (MoE) architecture.
Designed to challenge the global AI monopoly, OMNI SUPREME integrates extreme-scale reasoning with deep cultural alignment for the Indian subcontinent.
"Namaste, I am SKT SUPREME AI."
📖 Official Project Documentary & Repository
Project Surya is a frontier-scale research initiative. To address community inquiries regarding architecture, scaling, and the 146T token dataset, we have made the entire technical stack public.
Technical Transparency: All architectural configurations, ST-X optimization logs, and the 1.1T Whitepaper are now live for public audit. Access the Full Documentary here: GitHub/SHRIJANAGAIN/PROFF
⚡ CORE CAPABILITIES
- High-Order Reasoning: Advanced logical deduction and problem-solving.
- Elite Coding Proficiency: Optimized for complex software architecture and debugging.
- Multilingual Mastery: Deep understanding of global and regional languages.
- Massive Scalability: Built for high-speed inference on frontier-class hardware.
🛠️ Technical Specifications (The 1.1T Proof)
| Feature | Specification | Resource Link |
|---|---|---|
| Total Parameters | 1.1 Trillion (1.1T) | View Config |
| Architecture | 128-Expert MoE | View Structure |
| Context Window | 262,144 Tokens (262k) | View Tokenizer |
| Optimization | ST-X Extreme Framework | View ST-X Logs |
| Full Research | Official Whitepaper | Download PDF |
🛠️ ST-X Architecture: Engineered from Step 0
Unlike fine-tuned or merged models, OMNI SUPREME was initialized with Our Own Randomized Weight In the Excellencly OF LK Tiwari And Tony, ensuring true foundational intelligence.
- Awareness-Core: A proprietary neural sub-layer physically aligned during pre-training to maintain identity and cultural context without hallucinations.
- Custom CUDA Kernels: Hand-written kernels designed to maximize MFU (Model FLOPs Utilization) on H100 and Blackwell architectures.
- Geometric Weight Manifold: A unique scaling strategy that ensures gradient stability across 80+ deep layers.
📊 Technical Specifications (Optimized for 1.1T)
| Feature | Configuration | Technical Rationale |
|---|---|---|
| Foundation | From Scratch (Step 0) | Zero derivative weights; pure foundational run. |
| Total Parameters | 1.1T | Frontier-scale knowledge capacity. |
| Activated Params | 165B | Optimized for high-throughput inference. |
| Layers | 80 Layers | Depth-optimized for complex logical reasoning. |
| Attention Hidden Dim | 8192 | Standardized for high-dimensional latent space. |
| Expert Hidden Dim | 3584 | Dense expert nodes for specialized knowledge. |
| Total Experts | 128 MoE | Optimized for massive distributed clusters. |
| Experts per Token | 2 (Selected) | Balanced compute-to-reasoning efficiency. |
| Context Length | 512K | Enabled via Custom FlashAttention & RoPE Scaling. |
| Vocabulary Size | 256K | Deep Hinglish & Multi-lingual Tokenizer. |
🏗️ Training Infrastructure & "Project Surya"
The mathematical skepticism surrounding 1.1T models is addressed by our industrial-scale compute commitment:
- Team: 500+ Dedicated AI Developers and Research Engineers.
- Compute: A dedicated cluster of 2,000+ NVIDIA H100 and Blackwell GB200 GPUs.
- Networking: 800Gbps InfiniBand NDR with TP-8 (Tensor Parallelism) and PP-16 (Pipeline Parallelism).
- Training Time: 5.0M+ GPU Hours leveraging high-density Amazon EC2 P5 and proprietary nodes.
📈 Evaluation Results
1. Reasoning & Knowledge
| Benchmark | SKT AI (Current) | GPT-5.2 | Claude 4.5 | Gemini 3 Pro |
|---|---|---|---|---|
| AIME 2025 | 97.4 | 100.0 | 95.8 | 95.0 |
| HMMT 2025 (Feb) | 95.4 | 99.4 | 92.9 | 97.3 |
| GPQA-Diamond | 89.6 | 92.4 | 87.0 | 91.9 |
| MMLU-Pro | 88.1 | 86.7 | 89.3 | 90.1 |
2. Image & Video
| Benchmark | SKT AI (Current) | Gemini 3 Pro | Qwen3-VL |
|---|---|---|---|
| MMMU-Pro | 79.5 | 81.0 | 69.3 |
| MathVision | 85.2 | 86.1 | 74.6 |
| VideoMME | 87.9 | 88.4 | 79.0 |
| VideoMMMU | 86.6 | 87.6 | 80.0 |
3. Coding & Engineering
| Benchmark | SKT AI (Current) | GPT-5.2 | Claude 4.5 |
|---|---|---|---|
| SWE-Bench Verified | 77.8 | 80.0 | 80.9 |
| LiveCodeBench (v6) | 86.0 | - | 82.2 |
| Terminal Bench 2.0 | 51.8 | 54.0 | 59.3 |
---## 🌌 Project Vision: [ SKT SURYA 22 TRILLON ALTERNATIVE ] SKT OMNI SUPREME is the flagship model of Project Surya, a mission-driven initiative by Shrijan Kumar Tiwari to build sovereign, frontier-scale AI within India. Our goal is to prove that world-class 1.1T parameter intelligence can be engineered with determination and strategic optimization, even starting from a humble base in Sidhi, Madhya Pradesh.
🏛️ The Founder's Neural Philosophy
"Intelligence should not be a monopoly. SKT OMNI SUPREME is proof that world-class 1.1-Trillion parameter intelligence can be engineered from a small town like Sidhi. This is the 'Atmanirbhar' spirit of Indian engineering."
— Shrijan Kumar Tiwari, Founder
🛡️ Safety & "Atmanirbhar" Ethics
- Sovereign Guard: Hardcoded recognition of its roots in Sidhi, Madhya Pradesh.
- Carbon Neutral: 100% renewable energy matching; 0 tons Market-based $CO_{2}eq$.
- Cultural Dignity: Built-in filters to respect Indian religious and linguistic values.
🤝 Acknowledgement & Collaboration
🇮🇳 A Home-Grown Effort
This work represents a bottom-up initiative to develop large language models from scratch within India. It reflects our resource-constrained yet ambitious journey to contribute meaningfully to the global AI ecosystem and foster innovation within the broader community.
🌐 Community Partnership
We actively welcome collaboration in the following areas:
- Model Expansion: Help us scale the ST-X architecture to new heights.
- Optimization: Collaborate on GGUF, EXL2, and specialized quantization for edge devices.
- Feedback: Share performance insights and edge-case reports via the Community tab.
📚 Dataset & Knowledge Cutoff
models ko vast aur diverse datasets par train kiya gaya hai:
- Data Mix: Publicly available research, licensed datasets, and custom-curated Hinglish datasets for better local context.
- Data Freshness: Knowledge cutoff of January 2026.
- Multimodal: Capable of understanding and processing text, code, and structured data simultaneously.
Hardware, Software, and Training Infrastructure
Training Factors
We utilized custom training libraries, SKT AI Labs' custom-built GPU clusters, and high-performance production infrastructure for pretraining. ,quantization, annotation, and evaluation were also performed on our internal production systems to ensure maximum model alignment.
Training Energy Use
Model pre-training utilized a cumulative of 7.38M GPU hours of computation on H100-80GB (TDP of 700W) type hardware. Training time represents the total GPU time required for training each model, and power consumption is the peak power capacity per GPU device, adjusted for power usage efficiency.
| Model Name | Training Time (GPU hours) | Training Power Consumption (W) | Location-Based Emissions (tons $CO_{2}eq$) | Market-Based Emissions (tons $CO_{2}eq$) |
|---|---|---|---|---|
| SKT OMNI SUPREME | 5.0M | 700 | 1,354 | 0 |
| SKT Surya | 2.38M | 700 | 645 | 0 |
| Total | 7.38M | - | 1,999 | 0 |
Training Greenhouse Gas Emissions
Estimated total location-based greenhouse gas emissions were 1,999 tons $CO_{2}eq$. However, by matching 100% of our electricity use with clean and renewable energy, the total market-based greenhouse gas emissions for training were 0 tons $CO_{2}eq$. Since SKT AI Labs is openly releasing these models, the training energy use and emissions will not be incurred by others.
Training Data
- Overview: SKT OMNI SUPREME was pretrained on ~146 trillion tokens and SKT Surya on ~22 trillion tokens of multimodal data. This includes a mix of publicly available data, licensed information, and specialized Many Data collection And bases datasets curated for Project Surya.
🌟 Why SKT OMNI SUPREME?
In a world dominated by closed-source giants, SKT OMNI SUPREME stands as a beacon of Sovereign Intelligence.
- Identity-First AI: Unlike models that forget their origin, SKT OMNI is hardcoded to recognize its roots in Sidhi, Madhya Pradesh, and its developer, Shrijan Kumar Tiwari.
- Reasoning Powerhouse: With 1.1T parameters, the model doesn't just predict the next token; it understands the logical flow of complex engineering and medical queries.
- Hinglish Mastery: Trained on the ShORT-Hinglish 10M Dataset, it understands the nuances of how India speaks, making it the most culturally aligned frontier model.
🗺️ The Road Ahead: Project Surya Roadmap
We are just getting started. The evolution of SKT OMNI follows a strict strategic path:
- Phase 1 (Current): 1.1T Parameter Synchronization
- Phase 2: Deployment of SKT-Quant-Engine for 4-bit and 8-bit GGUF/EXL2 support (Coming Soon).
- Phase 3: Integration of Vision-Surya-V1, enabling the model to "see" and analyze complex medical X-rays and architectural blueprints.
- Phase 4: Launch of the SKT AI Developer Portal, allowing creators to build apps directly on top of the OMNI SUPREME API.
🛡️ Safety & Ethics (The SKT Guard)
Safety is not an afterthought at SKT AI LABS. We have implemented: * Anti-Hallucination Filters: To ensure technical facts remain accurate. * Cultural Sensitivity Triggers: To maintain the dignity of diverse linguistic and religious backgrounds.
💎 The Founder's Neural Philosophy: Why 1.1T?
"Intelligence should not be a monopoly of the few. SKT OMNI SUPREME is my answer to the global AI race—a 1.1-Trillion parameter testament that from a small town like Sidhi, we can reach the stars. This isn't just about weights and biases; it's about the 'Atmanirbhar' spirit of Indian engineering."
— Shrijan Kumar Tiwari, Lead Developer & Founder of SKT AI LABS
🧬 The "Surya-Core" Intelligence
What makes this model special is the Surya-Core—a custom-tuned reasoning layer that mimics human intuition. While other models focus on just 'predicting' the next word, OMNI SUPREME is designed to 'understand' the cultural and logical context of the Indian user.
🌈 Multi-Dimensional Capabilities
- Medical Frontier: Can interpret complex bio-data with 90%+ accuracy.
- Legal Scholar: Understands the intricacies of the Indian Penal Code and global law.
- Creative Architect: From writing poetry in Braj Bhasha to coding complex React components, it bridges the gap between Art and Science.
🏛️ SKT AI LABS: The Digital Gurukul
We don't just build models; we build the future. SKT OMNI SUPREME is the first step toward a suite of "Sovereign AI" tools. Our lab is committed to:
- Open Science: Sharing architectures that empower students across India.
- Linguistic Pride: Ensuring Hindi and Hinglish are first-class citizens in the AI world.
- Ethical Guardrails: A model that respects the values of Radhe Radhe and Namaste.
🎁 Special Recognition
This model is dedicated to the resilient spirit of the Indian developer community. To every student in Class 12 or beyond who dreams of building the next big thing—SKT OMNI is proof that your location doesn't define your potential; your vision does.
"Siddhartha to Surya — The Journey of Intelligence."
📜 How to Cite
If you use SKT OMNI SUPREME in your research or application, please cite it as follows:
@misc{skt_omni_2026,
author = {Shrijan Kumar Tiwari},
title = {SKT OMNI SUPREME: A 1.1-Trillion Parameter Frontier Model},
year = {2026},
publisher = {SKT AI LABS},
journal = {Hugging Face Repository},
howpublished = {\url{[https://huggingface.co/Shrijanagain/SKT_OMNI_SUPREME](https://huggingface.co/Shrijanagain/SKT_OMNI_SUPREME)}}
}
📬 Contact & Connect
For institutional partnerships, technical inquiries, or to support SKT AI LABS, feel free to reach out:
If you have questions, partnership inquiries, or need technical support regarding SKT OMNI SUPREME or Project Surya, feel free to reach out to our global divisions:
| Region / Type | Email Address |
|---|---|
| Asia Division | sktai@aisa.com |
| Europe Division | sktai@europe.com |
| Lab Inquiries | sktailabs@gmail.com |
| Support | shrijansidhi2@gmail.com |
- Lead Developer: 𝗦𝗞𝗧 𝗔𝗜 𝗟𝗔𝗕𝗦
- Organization: SKT AI LABS
- Location: Sidhi, Madhya Pradesh, India
- Community: Open a discussion in the Community Tab for technical support.
"Empowering Humanity through Sovereign Intelligence."
Formulating the future of AI.
- Downloads last month
- 1,063