Local LLMs for reverse engineering
Models for decompiling binaries, analysing firmware and malware, and reading Android apps offline, with the memory each needs and the Ghidra, IDA Pro, Binary Ninja, radare2 and JADX plugins that use them.
Step 1
Tell us your hardware first
Then you will see how many of these 18 LLMs run on it, and can filter the list by how well each one runs.
Two kinds of model are here. Decompiler models such as LLM4Decompile turn assembly or Ghidra’s pseudo-code back into readable C; run them on their own in llama.cpp or LM Studio. General models that the plugin makers recommend, plus Cisco’s security-tuned Foundation-Sec, explain code, rename functions and drive Ghidra, IDA Pro, Binary Ninja, radare2 or JADX through a reverse-engineering plugin. Only analyse software you are allowed to.
| Model | Made by | Arch | Parameters | Max context | Quality tier | Released | Runs in |
|---|---|---|---|---|---|---|---|
| Qwen3.5 27B | Alibaba | Dense | 28B | 256K | 93 | 2026-02 | +17 |
| Qwen3.5 35B-A3B | Alibaba | MoE | 36B (3.0B active) | 256K | 93 | 2026-02 | +17 |
| DeepSeek R1 (671B) | DeepSeek | MoE | 671B (37.0B active) | 128K | 92 | 2025-01 | +18 |
| GLM-4.7 Flash (30B-A3B) | Zhipu AI | MoE | 31B (3.0B active) | 198K | 88 | 2026-01 | +17 |
| Devstral 2 (123B) | Mistral AI | Dense | 125B | 256K | 86 | 2025-12 | +17 |
| gpt-oss 120B | OpenAI | MoE | 117B (5.1B active) | 128K | 84 | 2025-08 | +17 |
| Qwen3 Coder 30B-A3B | Alibaba | MoE | 31B (3.3B active) | 256K | 78 | 2025-07 | +17 |
| DeepSeek V3 (671B) | DeepSeek | MoE | 671B (37.0B active) | 128K | 75 | 2024-12 | +18 |
| Qwen3 30B-A3B | Alibaba | MoE | 31B (3.3B active) | 32K | 74 | 2025-04 | +18 |
| Gemma 3 27B IT | Dense | 27B | 128K | 72 | 2025-03 | +18 | |
| gpt-oss 20B | OpenAI | MoE | 21B (3.6B active) | 128K | 70 | 2025-08 | +17 |
| Qwen3 8B | Alibaba | Dense | 8.2B | 32K | 60 | 2025-04 | +18 |
| LLM4Decompile 22B V2 | LLM4Binary | Dense | 22B | 32K | 58 | 2024-06 | +15 |
| Foundation-Sec 8B Reasoning | Cisco Foundation AI | Dense | 8.0B | 128K | 52 | 2025-11 | +14 |
| Foundation-Sec 1.1 8B Instruct | Cisco Foundation AI | Dense | 8.0B | 128K | 50 | 2025-11 | +14 |
| LLM4Decompile 9B V2 | LLM4Binary | Dense | 8.8B | 128K | 38 | 2024-09 | +15 |
| LLM4Decompile 6.7B V2 | LLM4Binary | Dense | 6.7B | 8K | 32 | 2024-06 | +15 |
| LLM4Decompile 1.3B V2 | LLM4Binary | Dense | 1.4B | 16K | 12 | 2024-06 | +15 |
Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.