Can I run it? / Android phones
Best LLMs to run on Android phones
Updated November 2026
As of November 2026, the best model to run on Android phones is Nanbeige4.2 3B by Nanbeige (Kanzhun) (Small LM Index 67.2), at about 12 tokens/s with 4-bit weights. For faster replies, LFM3 1.7B (61.2) runs at about 29 tokens/s.
Flagship Android phones have 12–16 GB of memory and mid-range ones 8 GB. Apps get about half of it, and phones slow down as they heat up, so small models work best. About 6.0 GB is usable for a model, at 77 GB/s. The figures below assume 4-bit weights and a 4K-token context, and only include models that generate at least 5 tokens per second.
Models that run well on Android phones
The 25 best of the 38 models that run at 5+ tokens/s, ranked by the Small LM Index.
| # | Model | Index | Memory | Speed |
|---|---|---|---|---|
| 1 | Nanbeige4.2 3B Nanbeige (Kanzhun) · 4-bit · Apache-2.0 | 67.2 | 3.5 GB | ≈ 12 tok/s |
| 2 | Spark-X2.6 4B SparkLLM (XHToken) · 4-bit · Apache-2.0 | 66.5 | 3.3 GB | ≈ 12 tok/s |
| 3 | Spark-X2.5 4B SparkLLM (XHToken) · 4-bit · Apache-2.0 | 65.9 | 3.3 GB | ≈ 12 tok/s |
| 4 | LFM3 1.7B Liquid AI · 4-bit · LFM Open License | 61.2 | 1.6 GB | ≈ 29 tok/s |
| 5 | MiniCPM5 2B OpenBMB (MiniCPM) · 4-bit · Apache-2.0 | 60.9 | 2.3 GB | ≈ 20 tok/s |
| 6 | Qwen3 4B Thinking 2507 Alibaba (Qwen) · 4-bit · Apache-2.0 | 59.8 | 3.6 GB | ≈ 13 tok/s |
| 7 | Qwen3.5 4B Alibaba (Qwen) · 4-bit · Apache-2.0 | 59.7 | 3.1 GB | ≈ 13 tok/s |
| 8 | Nemotron Labs Dory 1.0 4B NVIDIA · 4-bit · OpenMDW-1.1 | 58.6 | 3.3 GB | ≈ 13 tok/s |
| 9 | Falcon H1R 7B TII (Falcon) · 4-bit · Falcon LLM License | 58.6 | 5.4 GB | ≈ 7 tok/s |
| 10 | Granite 4.2 3B IBM (Granite) · 4-bit · Apache-2.0 | 57.3 | 2.7 GB | ≈ 17 tok/s |
| 11 | Hunyuan 7B Instruct Tencent (Hunyuan) · 4-bit · Tencent Hunyuan Community | 55.8 | 5.7 GB | ≈ 7 tok/s |
| 12 | Ling 3.0 Tiny InclusionAI (Ling) · 4-bit · MIT | 54.9 | 5.6 GB | ≈ 18 tok/s |
| 13 | Ministral 3 3B Mistral AI · 4-bit · Apache-2.0 | 53.8 | 2.8 GB | ≈ 17 tok/s |
| 14 | Spark-X2.6 1.7B SparkLLM (XHToken) · 4-bit · Apache-2.0 | 52.3 | 1.7 GB | ≈ 30 tok/s |
| 15 | Edge0 8B-A1B (preview) Edge0 · 4-bit · Apache-2.0 | 50.1 | 5.6 GB | ≈ 18 tok/s |
| 16 | MiMo-7B-RL Xiaomi (MiMo) · 4-bit · MIT | 45.8 | 6.0 GB | ≈ 7 tok/s |
| 17 | DeepSeek-R1-Distill-Qwen-7B DeepSeek · 4-bit · MIT | 45.6 | 5.5 GB | ≈ 7 tok/s |
| 18 | LFM2.5 2.6B Liquid AI · 4-bit · LFM Open License | 44.7 | 2.3 GB | ≈ 19 tok/s |
| 19 | EXAONE 4.0 1.2B LG AI Research (EXAONE) · 4-bit · EXAONE License (non-commercial) | 42.5 | 1.6 GB | ≈ 40 tok/s |
| 20 | Gemma 4 E4B Google · 4-bit · Apache-2.0 | 40.9 | 5.6 GB | ≈ 11 tok/s |
| 21 | LFM2.5 8B-A1B Liquid AI · 4-bit · LFM Open License | 40.1 | 5.9 GB | ≈ 25 tok/s |
| 22 | Gemma 4 E2B Google · 4-bit · Apache-2.0 | 37.4 | 3.7 GB | ≈ 22 tok/s |
| 23 | SmolLM3 3B Hugging Face (SmolLM) · 4-bit · Apache-2.0 | 36.8 | 2.7 GB | ≈ 16 tok/s |
| 24 | Qwen3.5 2B Alibaba (Qwen) · 4-bit · Apache-2.0 | 35.9 | 2.0 GB | ≈ 22 tok/s |
| 25 | LFM2.5 1.2B Thinking Liquid AI · 4-bit · LFM Open License | 34.7 | 1.3 GB | ≈ 43 tok/s |
Change the context length or quantization, or see models that only just fit, in Can I run it? for this device.
How to run Nanbeige4.2 3B on Android phones
- Install an app. PocketPal AI (free, open source) downloads GGUF models from Hugging Face; Google AI Edge Gallery runs Gemma and other LiteRT models on the phone's GPU.
- Load Nanbeige4.2 3B. In PocketPal, search Hugging Face for bartowski/Nanbeige_Nanbeige4.2-3B-GGUF and pick the Q4_K_M file.
- Build it into an app. For developers: llama.cpp, LiteRT-LM or ExecuTorch.
Every model page lists the same commands for its own files. Nanbeige4.2 3B's page also has a memory table for other context lengths.
Questions
What is the best LLM for Android phones?
As of November 2026, the best model to run on Android phones is Nanbeige4.2 3B by Nanbeige (Kanzhun) (Small LM Index 67.2), at about 12 tokens/s with 4-bit weights. For faster replies, LFM3 1.7B (61.2) runs at about 29 tokens/s.
What is the biggest model that runs on Android phones?
LFM2.5 8B-A1B (8.5B parameters, 1B active) fits at 4-bit, needing 5.9 GB and generating about 25 tokens/s. Bigger models may fit at lower quality or run more slowly; try them in the Can I run it? tool.
How much memory can a model use on Android phones?
About 6.0 GB of the 12 GB. Flagship Android phones have 12–16 GB of memory and mid-range ones 8 GB. Apps get about half of it, and phones slow down as they heat up, so small models work best.
How accurate are these speeds?
They are estimates from memory bandwidth (77 GB/s here), which limits how fast a model generates text, with a typical efficiency for this kind of hardware. Expect real speeds within about a third either way. Long prompts take longer to read before the first word appears.