Orion-9B-Agentic-CodeCore-Merge

Orion-9B-Agentic-CodeCore-Merge is a merged 9B-parameter coding and reasoning model designed for long-horizon coding tasks, agentic coding, and agentic reasoning. It is built by merging Qwen3.5-9B as the base model with MiMo-V2.6-Distill-Qwen-9B, OxCoder-9B, and Ornith-1.5-9B, combining their capabilities for code generation, multi-step problem solving, instruction following, and autonomous coding workflows. The model is intended for complex software-engineering tasks that require sustained reasoning across multiple steps, code understanding, modification, debugging, and tool-oriented agentic workflows. This model is experimental and may generate artifacts.

• GGUF: Orion-9B-Agentic-CodeCore-Merge-GGUF. Note: The Multi-Token Prediction (MTP) heads are not preserved in this format. The model runs as a standard single-token-per-step autoregressive decoder.

Merge Recipe

hf download XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B \
--local-dir ./models/MiMo-V2.6-Distill-Qwen-9B

hf download ornith-ai/Ornith-1.5-9B \
--local-dir ./models/Ornith-1.5-9B

hf download OrionLLM/OxCoder-9B \
--local-dir ./models/OxCoder-9B

hf download Qwen/Qwen3.5-9B \
--local-dir ./models/Qwen3.5-9B

python omnimergekit.py \
--base      models/Qwen3.5-9B --task-base ./models/Qwen3.5-9B \
--source    models/MiMo-V2.6-Distill-Qwen-9B \
--source    models/OxCoder-9B \
--source    models/Ornith-1.5-9B \
--weights   0.55,0.30,0.15 \
--method omnimerge_v2 --density 0.53 --darex-q 0.75 --seed 42 \
--no-auto-mlp-skip --skip-patterns visual.,mtp. \
--output prithivMLmods/Orion-9B-Agentic-CodeCore-Merge

omnimergekit

Tools for model merging, expert pruning, differential competence-map extraction, and GGUF quantization — omnimergekit