File size: 13,153 Bytes
15a8db8 a1d6528 15a8db8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 | ---
library_name: transformers
license: apache-2.0
pipeline_tag: text-generation
base_model:
- Qwen/Qwen3.5-9B
---
## Introduction
We are proud to introduce **OxCoder-9B**, our lightweight coding model for long-horizon tasks, agentic coding, and agentic reasoning. Despite featuring only 9 billion parameters, OxCoder-9B marks a substantial leap in agentic capabilities — particularly in **Agentic Terminal** and **Agentic Coding** tasks — punching far above its weight class and rivaling models many times its size. This release represents our strongest commitment yet to delivering frontier-level agentic performance in a compact, efficient, and openly accessible package.
- **Trained on Frontier Agent Traces:** Built from Fable-5.1 and GLM-5.3 agentic coding trajectories across Claude Code, OpenCode and Codex — OxCoder-9B has been distilled from some of the most capable agentic systems available, giving it a unique advantage in real-world software engineering scenarios.
- **262K Native Context:** Equipped with a full 262,144 token native context window, enabling the model to handle complex, multi-file codebases and long-horizon reasoning tasks with ease.
- **Error Recovery:** Learns read-before-write patterns, responds to LSP diagnostics, and applies minimal edit diffs instead of full rewrites — making it robust in iterative coding environments where precision matters.
- **Front-end Reasoning:** OxCoder-9B exhibits remarkably strong front-end reasoning capabilities for its size, demonstrating a deep understanding of UI logic, component architecture, and web-native patterns that is rare in sub-10B models.
## Benchmark
<table>
<tr>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;"> </th>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;">OxCoder-9B</th>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;">Ornith-1.5-9B</th>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;">Ornith-1.0-9B</th>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;">Qwen3.5-9B</th>
<th style="background: rgba(128,128,128,0.1); text-align: center; border: hidden;">Gemma-4-31B</th>
</tr>
<tr>
<td align="center" colspan="6" style="background: linear-gradient(90deg, rgba(124,58,237,0.45) 0%, rgba(99,102,241,0.42) 50%, rgba(59,130,246,0.45) 100%); font-weight: bold; height:32px; padding-top:2px; padding-bottom:2px;"><i>Coding</i></td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Agentic terminal coding</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">Terminal-Bench 2.1 (Terminus-2)</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(124,58,237,0.16) 0%, rgba(118,65,238,0.16) 100%); vertical-align:middle;"><b>49.6</b></td>
<td align="center" style="vertical-align:middle;">46.2</td>
<td align="center" style="vertical-align:middle;">43.1</td>
<td align="center" style="vertical-align:middle;">21.3</td>
<td align="center" style="vertical-align:middle;">42.1</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Agentic terminal coding</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">Terminal-Bench 2.1 (Claude Code)</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(118,65,238,0.16) 0%, rgba(112,71,239,0.16) 100%); vertical-align:middle;"><b>50.8</b></td>
<td align="center" style="vertical-align:middle;">47.0</td>
<td align="center" style="vertical-align:middle;">40.6</td>
<td align="center" style="vertical-align:middle;">18.9</td>
<td align="center" style="vertical-align:middle;">—</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Agentic coding</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">SWE-bench Verified</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(112,71,239,0.16) 0%, rgba(106,78,239,0.16) 100%); vertical-align:middle;"><b>73.5</b></td>
<td align="center" style="vertical-align:middle;">70.6</td>
<td align="center" style="vertical-align:middle;">69.4</td>
<td align="center" style="vertical-align:middle;">53.2</td>
<td align="center" style="vertical-align:middle;">52.0</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Real-world software engineering</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">SWE-bench Pro</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(106,78,239,0.16) 0%, rgba(100,84,240,0.16) 100%); vertical-align:middle;"><b>49.1</b></td>
<td align="center" style="vertical-align:middle;">47.5</td>
<td align="center" style="vertical-align:middle;">42.9</td>
<td align="center" style="vertical-align:middle;">31.3</td>
<td align="center" style="vertical-align:middle;">35.7</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Repo-level code generation</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">NL2Repo</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(100,84,240,0.16) 0%, rgba(94,91,241,0.16) 100%); vertical-align:middle;"><b>36.2</b></td>
<td align="center" style="vertical-align:middle;">32.4</td>
<td align="center" style="vertical-align:middle;">27.2</td>
<td align="center" style="vertical-align:middle;">16.2</td>
<td align="center" style="vertical-align:middle;">15.5</td>
</tr>
<tr>
<td align="center" colspan="6" style="background: linear-gradient(90deg, rgba(124,58,237,0.45) 0%, rgba(99,102,241,0.42) 50%, rgba(59,130,246,0.45) 100%); font-weight: bold; height:32px; padding-top:2px; padding-bottom:2px;"><i>Reasoning</i></td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Expert-level reasoning</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">HLE (no tools)</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(94,91,241,0.16) 0%, rgba(89,97,242,0.16) 100%); vertical-align:middle;"><b>21.2</b></td>
<td align="center" style="vertical-align:middle;">20.2</td>
<td align="center" style="vertical-align:middle;">16.8</td>
<td align="center" style="vertical-align:middle;">14.7</td>
<td align="center" style="vertical-align:middle;">19.5</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Tool-augmented reasoning</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">HLE (with tools)</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(89,97,242,0.16) 0%, rgba(83,104,243,0.16) 100%); vertical-align:middle;"><b>32.8</b></td>
<td align="center" style="vertical-align:middle;">30.5</td>
<td align="center" style="vertical-align:middle;">26.4</td>
<td align="center" style="vertical-align:middle;">24.5</td>
<td align="center" style="vertical-align:middle;">26.5</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Scientific reasoning</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">GPQA Diamond</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(83,104,243,0.16) 0%, rgba(77,110,244,0.16) 100%); vertical-align:middle;"><b>86.9</b></td>
<td align="center" style="vertical-align:middle;">86.4</td>
<td align="center" style="vertical-align:middle;">82.5</td>
<td align="center" style="vertical-align:middle;">81.7</td>
<td align="center" style="vertical-align:middle;">84.3</td>
</tr>
<tr>
<td align="center" colspan="6" style="background: linear-gradient(90deg, rgba(124,58,237,0.45) 0%, rgba(99,102,241,0.42) 50%, rgba(59,130,246,0.45) 100%); font-weight: bold; height:32px; padding-top:2px; padding-bottom:2px;"><i>Agentic</i></td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Multi-tool orchestration</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">MCP-Atlas</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(77,110,244,0.16) 0%, rgba(71,117,244,0.16) 100%); vertical-align:middle;"><b>56.7</b></td>
<td align="center" style="vertical-align:middle;">54.2</td>
<td align="center" style="vertical-align:middle;">49.4</td>
<td align="center" style="vertical-align:middle;">46.8</td>
<td align="center" style="vertical-align:middle;">55.0</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Web browsing and research</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">BrowseComp</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(71,117,244,0.16) 0%, rgba(65,123,245,0.16) 100%); vertical-align:middle;"><b>57.4</b></td>
<td align="center" style="vertical-align:middle;">56.4</td>
<td align="center" style="vertical-align:middle;">44.8</td>
<td align="center" style="vertical-align:middle;">41.5</td>
<td align="center" style="vertical-align:middle;">—</td>
</tr>
<tr>
<td style="text-align:left; padding:9px 10px 9px 18px;"><div style="font-size:15px; font-weight:600; line-height:1.22;">Real-user agentic coding</div><div style="margin-top:4px; font-size:11px; font-weight:400; line-height:1.2; opacity:0.62;">ClawEval</div></td>
<td align="center" style="background: linear-gradient(135deg, rgba(65,123,245,0.16) 0%, rgba(59,130,246,0.16) 100%); vertical-align:middle;"><b>67.8</b></td>
<td align="center" style="vertical-align:middle;">66.5</td>
<td align="center" style="vertical-align:middle;">63.1</td>
<td align="center" style="vertical-align:middle;">53.2</td>
<td align="center" style="vertical-align:middle;">48.5</td>
</tr>
</table>
<p style="margin-top:12px;font-size:10px;opacity:0.7">
* All results reported for OxCoder-9B are averaged over five independent runs. A dash (—) means the score was not reported for that model.<br/>
* Terminal-Bench 2.1 (Terminus-2): We evaluate Terminal-Bench 2.1 using the Harbor/Terminus-2 framework with parser=json, temperature=1.0, top_p=1.0, and a 256K context window. Each run uses a 2-hour timeout with 32 CPU cores and 32GB RAM.<br/>
* Terminal-Bench 2.1 (Claude Code): We evaluate Terminal-Bench 2.1 using Claude Code 2.1.126 with parser=json, temperature=1.0, top_p=1.0.<br/>
* SWE-bench Verified and Pro: using the OpenHands harness with temp=1.0, top_p=0.95, 256K context window. Anti-hacking safeguards are applied throughout evaluation: Git history is removed from the local repository image to prevent access to prior solutions or commits; network access is disabled, preventing the model from retrieving external information or resources.<br/>
* NL2Repo: with temperature=1.0, top_p=1.0, 256K context, 48K output. Access to the specified GitHub repositories and pip packages is blocked to prevent reward hacking.<br/>
* HLE: Evaluated using GLM-5.3 as the judge model.<br/>
* MCP-Atlas: Evaluated using GLM-5.3 as the judge model.<br/>
* ClawEval: temp=0.6 and 256K context.<br/>
* Baseline scores for Ornith-1.5-9B, Ornith-1.0-9B, Qwen3.5-9B and Gemma-4-31B are taken from the Ornith-1.5-9B model card and were produced under that team's evaluation settings, which differ from ours on some benchmarks (context window, timeouts and judge model).<br/>
</p>
---
<div align="center">
**OxCoder-9B** is developed by **[OrionLLM](https://huggingface.co/OrionLLM)** and released under the Apache 2.0 License.
</div> |