Models · on your own machine
Local LLM for reverse engineering
The local LLM for reverse engineering in an air-gapped lab is Qwen 3.8 27B Abliterated. It is about 17 GB at 4-bit and fits one 24 GB GPU or a 32 GB Mac.
Models that fit
| Model | Parameters | Size at 4-bit | Runs on | License |
|---|---|---|---|---|
| Qwen 3.8 27B Abliteratedhuihui-ai/Huihui-Qwen3.8-27B-abliterated | 27.8B | about 17 GB | one 24 GB GPU or a 32 GB Mac | Apache 2.0 |
| Qwen 3.6 35B-A3B Abliteratedhuihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated | 36B total, 3B active | about 22 GB | one 24 GB GPU | Apache 2.0 |
| Gemma 4 31B Hereticcoder3101/gemma-4-31B-it-heretic | 31.3B | about 19 GB | one 24 GB GPU | Apache 2.0 |
| Llama 3.1 8B Abliteratedmlabonne/Meta-Llama-3.1-8B-Instruct-abliterated | 8B | about 5 GB | an 8 GB GPU or a laptop | Llama 3.1 Community License |
From each model's Hugging Face page, read November 18, 2026. Sizes at 4-bit are estimates.
Qwen 3.8 27B Abliterated. Its GGUF build had over 4.3 million downloads in the last month. Qwen 3.6 35B-A3B Abliterated. Its MTP GGUF build had over 5 million downloads in the last month. Gemma 4 31B Heretic. Abliterated with the Heretic tool. Llama 3.1 8B Abliterated. Published by mlabonne, whose 2024 write-up made abliteration popular.
When to run locally
These models are 8B to 36B parameters. Kimi K3.1 Pro is 2,780B and writes up to 943,718 tokens.
Plugins
| Tool | Local servers in its README | Setting |
|---|---|---|
| Gepetto (IDA Pro) | Ollama, LM Studio | EditGepetto, or BASE_URL in config.ini |
| r2ai (radare2) | Ollama, LM Studio | r2ai -e api=openai and r2ai -e baseurl=http://LM-STUDIO-IP:PORT |
| GhidrAssist (Ghidra) | Ollama, LM Studio, Open-WebUI | Provider type Ollama, URL http://localhost:11434 |
| IDAssist (IDA Pro) | Ollama | Type ollama, URL http://localhost:11434 |
| BinAssist (Binary Ninja) | Ollama, LM Studio, Open-WebUI | API Host http://localhost:11434 |
An MCP agent can use the same address. See the MCP setup page.
Trade-off
A 27B model names simple functions and drops details in long ones. Rerun the hard ones on a larger model.