Advanced coding
Cleaner generation and refactoring across C++, Rust, Python, and TypeScript, with deep architectural comprehension of large codebases.
New Flagship · Successor to Sylor S2 Pro
Sylor S2.1 Pro.
The second-generation flagship successor: 744B parameters in a MoE architecture with 40B active per token, 1M native context, and major leaps in reasoning and coding.
- Total parameters
- 744B
- Active parameters
- 40B
- Native context
- 1M tokens
The family
What Sylor is.
Sylor is the family of AI models Orzatty trains and serves from sylor.orzatty.com. The name splits generation and profile: S2 Flash for fast inference (284B MoE, 13B active), S2 Pro and S2.1 Pro as the flagships (744B MoE, 40B active) and S3 Preview as the third generation in closed preview (2.8T MoE, 104B active). All of them share a native 1M token context and the same method: continual pre-training and SFT on top of a base model.
Open weights · Coming soon
Open GGUF weights in preparation.
Quantized weights for Sylor S2.1 Pro will be published in GGUF format on Hugging Face (date TBA). In the meantime, you can try the model directly on sylor.orzatty.com or download weights for its predecessor Sylor S2 Pro.
Specifications
Technical sheet.
Sylor S2.1 Pro consolidates the Mixture-of-Experts architecture with 744B total parameters, 40B active, and direct enhancements in agentic stability and software synthesis.
Total parameters 744B
Active parameters 40B
Architecture Mixture-of-Experts
Native context 1M tokens
Open format GGUF (in preparation)
Predecessor Sylor S2 Pro (GGUF available)
Status Available on sylor.orzatty.com
Training Massive CPT and SFT on a base model
Hardware NVIDIA B300 cluster
The sylor.orzatty.com platform currently exposes a 256K limit.
Key improvements
Evolution from S2 Pro to S2.1 Pro.
S2.1 Pro advances second-generation strengths with sharper agentic execution, greater synthetic consistency, and refined multi-step reasoning.
Reasoning & autonomy
Greater resilience in looped tool execution, logical deduction, and systematic bug troubleshooting.
1M context
One-million token native context window to process entire repositories and dense specifications in a single prompt session.
Inference efficiency
40B active parameters per token: flagship-grade capability with optimized latency for distributed deployments.
Frequently asked questions
Sylor, in short.
What is Sylor?
Orzatty’s family of AI models. Today it is made up of S2 Flash, S2 Pro, S2.1 Pro and S3 Preview. All of them run on sylor.orzatty.com, which currently exposes a 256K context limit.
Which Sylor model is the best?
Sylor S2.1 Pro is the flagship: a 744B MoE with 40B active parameters per token and a native 1M context. Sylor S3 Preview is the largest (2.8T with 104B active) and sits in closed preview. For fast inference there is S2 Flash, with 284B and 13B active.
Does Sylor have open weights?
Partly. The GGUF weights for Sylor S2 Pro have been published on Hugging Face since September 2, 2026. Those for S2.1 Pro are in preparation, with no published date. S3 Preview only runs in closed preview.
Where can I try Sylor?
On https://sylor.orzatty.com. The family API will arrive on Orzatty AI Studio (https://aistudio.orzatty.com).
Experience Sylor S2.1 Pro.
Access the model on sylor.orzatty.com or reach out to our infrastructure team.