Ornith 1.5 35B (A3B)
36B parameter open-weight model. Sparse mixture-of-experts - 35.95B total across 256 experts with 8 routed per token (about 3B active), on the Qwen3.5 MoE architecture. 256K context, image input, MIT. Same declared derivative lineage as the 9B (model card read 2026-08-30). The ornith-1.5:35b tag was verified against the Ollama registry manifest, which serves a 21.71 GB Q4_K_M plus a 0.90 GB vision projector (checked 2026-08-30). This project measured 35.1 tok/s on an RTX 3080 (10 GB) with the model split 64% CPU / 36% GPU (2026-08-21, docs/lab/radar-2026-08-21-model-validation.md) - a mixed-offload result on a card far smaller than the weights, not evidence that it fits in 10 GB of VRAM. Sparsity is what keeps a heavily split run usable.
Ornith AI · Ornith
Model overview
Ornith 1.5 35B (A3B) is the sparse mixture-of-experts member of the Ornith 1.5 family: 35.95B total parameters across 256 experts with roughly 3B active per token, a 256K context window and image input. Because only a small fraction of the weights is active per token, it stays responsive even when most of the model sits in system RAM rather than VRAM, which is how this project measured it. Treat the Q4_K_M figure as the full weight footprint, not a VRAM requirement a smaller card satisfies.
Editorial review
VRAM figures are empirical estimates. Actual usage varies by runtime, context length, and system configuration. Verify on your specific hardware before production use.
Will Ornith 1.5 35B (A3B) run on your machine?
Ornith 1.5 35B (A3B) is 36B parameters and needs 23.5 GB of VRAM at Q4_K_M — 22 GB of weights plus 1.5 GB of runtime overhead for the inference server itself.
VRAM by quantization
| Quantization | Weights | Needs (with overhead) | Quality |
|---|---|---|---|
| Q4_K_M | 22 GB | 23.5 GB | good |
Fit on common hardware at Q4_K_M
| Hardware | Memory the model can use | System RAM | Verdict |
|---|---|---|---|
| CPU Only | None (CPU only) | 16 GB | Too large |
| RTX 4060 Laptop | 8 GB | 16 GB | Too large |
| RTX 3060 (12GB) | 12 GB | 32 GB | Too large |
| RTX 4060 Ti (16GB) | 16 GB | 32 GB | Too large |
| RTX 3090 | 24 GB | 64 GB | Tight |
| Apple Silicon (Unified Memory) 36 GB | 27 GB of 36 GB | 36 GB | Comfortable |
| RTX 5090 | 32 GB | 64 GB | Comfortable |
| Apple Silicon (Unified Memory) 48 GB | 36 GB of 48 GB | 48 GB | Comfortable |
Comfortable means VRAM clears the requirement by 2 GB or more. Tight means it covers the requirement with no margin. CPU offload means the model does not fit in VRAM but system RAM is at least 1.6× the weights, so it will run at reduced speed — expect roughly 1–5 tokens per second. Figures are weights plus a fixed runtime overhead and exclude KV-cache growth, which scales with context length.
Apple Silicon shares one pool of memory between the system and the GPU, so a model cannot use all of it. These rows apply the same 75% usable fraction the Compatibility Checker uses, which is why a 36 GB Mac is graded on less than 36 GB.
Need more hardware for Ornith 1.5 35B (A3B)? Open the PC Builder for the 30B / 32B tier →
Ready to run this model locally?
Find a compatible interface in our Local AI Tools directory →