General Q&A · Coding
Ornith-1.5-35B-A3B
A 35B mixture-of-experts model for text reasoning, coding, and agent tasks. The PINVOU Model Store offers a single-device BF16 deployment that connects to PINVOU through a compatible API.
01 / Model overview
Model architecture & capabilities
Ornith-1.5-35B-A3B builds on the Qwen3.5 family through continued pretraining by Ornith AI and reinforcement learning for reasoning, coding, and tool use. Thinking mode is enabled by default for tasks involving multi-step reasoning or tools. The model configuration includes a vision encoder; the current Model Store deployment primarily targets text conversations and coding.
- Publisher / model
- Ornith AI · Ornith-1.5-35B-A3B
- Parameter count
- 35B total parameters; approx. 3B active per token
- Architecture
- Qwen3.5 MoE architecture; 40 layers, 256 routed experts, 8 selected per step, including shared experts
- Model context
- Configured limit of 262,144 tokens; store deployment reference uses 32K
- Input / output
- Text input with text / tool-call output; the model configuration also includes a vision encoder and image/video tokens
- Precision / download size
- BF16; approx. 71.9 GB (67.0 GiB), not runtime memory usage
- License
- MIT
02 / Use cases
Use cases
Knowledge Q&A and multi-step reasoning
Answer questions about written material, break down complex tasks, and assist with analysis and content organization.
Code development and bug fixing
Write and explain code, investigate errors, and help fix software issues.
Agent workflows with tools
Use an application that supports tool calling to coordinate queries, analysis, and actions across a multi-step task.
Plan model deployment and application integration
Evaluate device configurations and inference services for your model size, workloads, and concurrency needs.