OxCoder-9B-GGUF
OxCoder-9B is developed by OrionLLM, a lightweight 9-billion-parameter coding model built on Qwen3.5-9B for long-horizon agentic coding, terminal automation, and agentic reasoning, distilled from frontier agentic trajectories generated by Fable-5.1 and GLM-5.3 across Claude Code, OpenCode, and Codex environments. It ships with a full 262,144-token native context window for handling complex multi-file codebases, learns error-recovery behaviors like read-before-write patterns and LSP-diagnostic-driven minimal edit diffs rather than full rewrites, and exhibits notably strong front-end reasoning for a sub-10B model, with deep understanding of UI logic and component architecture. Despite its compact size, OxCoder-9B punches well above its weight class, outperforming Ornith-1.5-9B, Ornith-1.0-9B, Qwen3.5-9B, and even the larger Gemma-4-31B across nearly every reported benchmark — including 50.8 on Terminal-Bench 2.1 (Claude Code), 73.5 on SWE-bench Verified, 49.1 on SWE-bench Pro, 86.9 on GPQA Diamond, and 67.8 on ClawEval — reflecting particular strength in agentic terminal coding and real-world software engineering tasks relative to its parameter count. The model is released under the Apache 2.0 License.
Model Files
| File Name | Quant Type | File Size | File Link |
|---|---|---|---|
| OxCoder-9B.BF16.gguf | BF16 | 17.9 GB | Download |
| OxCoder-9B.Q3_K_L.gguf | Q3_K_L | 4.93 GB | Download |
| OxCoder-9B.Q3_K_M.gguf | Q3_K_M | 4.62 GB | Download |
| OxCoder-9B.Q3_K_S.gguf | Q3_K_S | 4.26 GB | Download |
| OxCoder-9B.Q4_0.gguf | Q4_0 | 5.31 GB | Download |
| OxCoder-9B.Q4_K_M.gguf | Q4_K_M | 5.63 GB | Download |
| OxCoder-9B.Q4_K_S.gguf | Q4_K_S | 5.35 GB | Download |
| OxCoder-9B.Q5_0.gguf | Q5_0 | 6.31 GB | Download |
| OxCoder-9B.Q5_K_M.gguf | Q5_K_M | 6.47 GB | Download |
| OxCoder-9B.Q5_K_S.gguf | Q5_K_S | 6.31 GB | Download |
| OxCoder-9B.mmproj-bf16.gguf | mmproj-bf16 | 922 MB | Download |
llama.cpp
LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp