State-of-the-Art Coding Agents: Post-trained on top of Qwen 3.5, achieving state-of-the-art performance among open-source models
Self-Improving Training Framework: Ornith-1.0 employs RL to learn to generate not only solution rollouts, but also the scaffold that drives those rollouts
35B total parameters with 8B active per token (256 experts, 8 active)
40-layer MoE architecture with sliding + full attention hybrid
262K context window
MIT License — globally accessible, no regional limitations

Architecture

Text model: Qwen3.5 MoE — 40 layers, 2048 hidden, 256 experts (8 active/token)
Vocabulary: 248,320 tokens

Quantization

Quantized from the BF16 safetensors using llama.cpp (build 537).

NVFP4 (NVIDIA FP4) uses 4-bit floating point quantization optimized for NVIDIA Blackwell GPUs.

File	Size	Description
`ornith-1.0-35b-nvfp4.gguf`	~18.4 GB	NVFP4 quantized model

bash

llama-server \
  -m ornith-1.0-35b-nvfp4.gguf \
  -ngl 99 \
  --host 0.0.0.0 \
  --port 8080

MIT