File size: 7,633 Bytes
88550ab fe4bc08 88550ab fe4bc08 88550ab fe4bc08 88550ab fe4bc08 88550ab d84656f 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 5728594 7290272 dbf990a 7290272 0696ea8 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 | ---
license: apache-2.0
base_model:
- Qwen/Qwen3.5-4B
pipeline_tag: image-text-to-text
tags:
- reasoning
- vision
- multimodal
- instruct
- chat
- coding
- math
- science
---
# GRM-2.5
<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/685ea8ff7b4139b6845ce395/sZK90f3KIE6JcF4Nrm_2g.png" alt="logo" width="500">
</p>
<div align="center">
<a href="https://huggingface.co/OrionLLM/GRM-2.5/" style="text-decoration: none;">
<img src="https://img.shields.io/badge/🤗-HuggingFace-FC926C?style=for-the-badge" alt="HuggingFace">
</a>
<a href="https://huggingface.co/collections/OrionLLM/grm-25" style="text-decoration: none;">
<img src="https://img.shields.io/badge/📚-Collection-3B82F6?style=for-the-badge" alt="Collection">
</a>
<a href="https://www.apache.org/licenses/LICENSE-2.0" style="text-decoration: none;">
<img src="https://img.shields.io/badge/📜-License-E343BD?style=for-the-badge" alt="License">
</a>
</div>
## 1. Introduction
GRM-2.5 is a **4B-parameter reasoning model** built for **general-purpose local AI**. It is designed to deliver strong performance across a wide range of tasks while remaining efficient and accessible for local inference.
The model is optimized for **structured reasoning**, helping it produce more accurate, coherent, and reliable responses on complex problems. GRM-2.5 aims to combine strong reasoning ability, practical usability, and efficient deployment in a compact form factor.
## 2. Key Capabilities
- **Strong Reasoning for Everyday and Advanced Tasks:** GRM-2.5 is built to handle both daily conversations and more demanding reasoning workloads with clarity and consistency.
- **Efficient Local Coding and Agentic Use:** Despite its compact size, the model is well suited for code generation, structured problem-solving, and local agent-style workflows.
- **Optimized for Local Deployment:** GRM-2.5 is designed for accessible inference across a broad range of hardware, making it a practical choice for users who want capable AI running locally.
## 3. Performance
GRM-2.5 is designed to be a highly capable option for **local AI use** across many scenarios. It performs well in **complex reasoning tasks, everyday chat, coding, and agentic workflows**, while maintaining the efficiency expected from a compact 4B model.
Its focus is not only raw capability, but also **practical intelligence**: strong reasoning, stable long-context behavior, and usability on consumer hardware.
<table>
<tr>
<th style="background: rgba(128,128,128,0.1); text-align: center;"> </th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">GRM-2.5-Plus (Closed)</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">GRM-2.5</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">GRM-2.5-Air</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">GRM-7B</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">GRM-1.5B</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>Knowledge & STEM</i></td>
</tr>
<tr>
<td align="center">MMLU-Pro</td>
<td align="center"><b>84.2</b></td>
<td align="center"><b>80.1</b></td>
<td align="center"><b>43.6</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
<tr>
<td align="center">GPQA Diamond</td>
<td align="center"><b>82.7</b></td>
<td align="center"><b>76.7</b></td>
<td align="center"><b>12.5</b></td>
<td align="center">53.7</td>
<td align="center">29.5</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>Instruction Following</i></td>
</tr>
<tr>
<td align="center">IFEval</td>
<td align="center"><b>91.8</b></td>
<td align="center"><b>90.2</b></td>
<td align="center"><b>44.5</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
<tr>
<td align="center">MultiChallenge</td>
<td align="center"><b>56.5</b></td>
<td align="center"><b>49.8</b></td>
<td align="center"><b>19.3</b></td>
<td align="center">--</td>
<td align="center">--</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 & Coding</i></td>
</tr>
<tr>
<td align="center">HMMT Feb 25</td>
<td align="center"><b>84.4</b></td>
<td align="center"><b>75.2</b></td>
<td align="center"><b>--</b></td>
<td align="center">42.7</td>
<td align="center">27.3</td>
</tr>
<tr>
<td align="center">HMMT Nov 25</td>
<td align="center"><b>83.2</b></td>
<td align="center"><b>77.2</b></td>
<td align="center"><b>--</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
<tr>
<td align="center">LiveCodeBench v6</td>
<td align="center"><b>67.2</b></td>
<td align="center"><b>56.9</b></td>
<td align="center"><b>--</b></td>
<td align="center">51.7</td>
<td align="center">39.4</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>Agent</i></td>
</tr>
<tr>
<td align="center">TAU2-Bench</td>
<td align="center"><b>80.5</b></td>
<td align="center"><b>80.2</b></td>
<td align="center"><b>11.6</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
<tr>
<td align="center">DeepPlanning</td>
<td align="center"><b>18.6</b></td>
<td align="center"><b>17.9</b></td>
<td align="center"><b>--</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
<tr>
<td align="center">OSWorld-Verified</td>
<td align="center"><b>42.4</b></td>
<td align="center"><b>36.0</b></td>
<td align="center"><b>--</b></td>
<td align="center">--</td>
<td align="center">--</td>
</tr>
</table>
## 4. Family
The GRM-2.5 family is available in various sizes to suit every case.
<table>
<tr>
<th style="background: rgba(128,128,128,0.1); text-align: center;">Model</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">Size</th>
<th style="background: rgba(128,128,128,0.1); text-align: center;">Domain</th>
</tr>
<tr>
<td align="center">GRM-2.5-Plus</td>
<td align="center">9B</td>
<td align="center">Closed model for research and agent purposes</td>
</tr>
<tr>
<td align="center">GRM-2.5</td>
<td align="center">4B</td>
<td align="center">Powerful on-device deployment for difficult tasks</td>
</tr>
<tr>
<td align="center">GRM-2.5-Air</td>
<td align="center">0.8B</td>
<td align="center">Any-device deployment for everyday chat</td>
</tr>
</table>
## 5. Architecture
GRM-2.5 is built on the Qwen3.5 architecture and is optimized for complex tasks, agent environments, and everyday chat.
GRM-2.5 applies the same principle to a stronger, larger foundation, resulting in a model that punches above its weight class on structured reasoning tasks while remaining deployable on consumer hardware.
---
GRM-2.5 is developed by [OrionLLM](https://huggingface.co/OrionLLM) and released under the Apache 2.0 License. |