Silkon 1T
1 trillion parameters. Unified intelligence.
Silkon 1T is our flagship model, trained on 10 trillion tokens of carefully curated data. It achieves state-of-the-art performance across reasoning, coding, mathematics, and multilingual understanding.
Specifications
Parameters
1 Trillion
Dense Transformer
Context Length
128K tokens
4096 max output
Training Data
10T tokens
Filtered + deduplicated
Architecture
Dense Transformer
RoPE, SwiGLU, RMSNorm
Tokenizer
128K vocab
Byte-level BPE
License
Research + Commercial
Contact for terms
Benchmarks
| Task | Silkon 1T | Baseline |
|---|---|---|
| MMLU | 87.2 | GPT-4: 86.4 |
| HumanEval | 92.1 | GPT-4: 89.0 |
| GSM8K | 94.8 | GPT-4: 92.0 |
| MATH | 78.3 | GPT-4: 76.6 |
| HumanEval+ | 89.5 | GPT-4: 85.4 |
| MBPP | 88.7 | GPT-4: 82.0 |
Capabilities
Use Cases
Research assistants
Accelerate scientific discovery with deep literature understanding and hypothesis generation.
Code review & generation
Production-grade code across 50+ languages with context-aware suggestions.
Enterprise automation
Deploy at scale with consistent quality for document processing, support, and analysis.
Multilingual applications
Native-quality understanding and generation across 100+ languages.
Ready to build with Silkon 1T?
Join the waitlist for early access and be among the first to deploy frontier AI.