Good and Small models for Mobile Devices
Try them out in
Privacy AI
on the App Store.
Privacy AI is a lightweight, serverless application. All tools - including web search, stock quotes, and Health analysis - run on-device, keeping data and actions fully private. It supports both local AI models and connections to your own OpenAI-compatible servers.
Qwen3 4B Q4
Qwen3 is the latest generation of Qwen series. It is a large language model with 1.7B parameters, optimized for mobile devices. It is capable of running functions on mobile devices and has been trained to follow instructions and generate long texts (32K tokens). It is more resilient to the diversity of system prompts, enhancing role-play implementation and condition-setting for chatbots. It has up to 128K tokens and can generate up to 32K tokens. It supports for over 100 languages, including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, Arabic, and more.
Model Intention: It is 4B of Qwen3 series that is excellent for summary, translation and MCP tool calling
Model Info URL: https://huggingface.co/Qwen/Qwen3-4B
Model License: License Info
Model Description: Qwen3 is the latest generation of Qwen series. It is a large language model with 1.7B parameters, optimized for mobile devices. It is capable of running functions on mobile devices and has been trained to follow instructions and generate long texts (32K tokens). It is more resilient to the diversity of system prompts, enhancing role-play implementation and condition-setting for chatbots. It has up to 128K tokens and can generate up to 32K tokens. It supports for over 100 languages, including Chinese, English, French, Spanish, Portuguese, German, Italian, Russian, Japanese, Korean, Vietnamese, Thai, Arabic, and more.
Developer: https://huggingface.co/Qwen
File Size: 2230 MB
Context Length: 1024 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
GLM Edge 4B Chat
GLM-4 is the latest generation of pre-trained models in the GLM-4 series launched by Zhipu AI. In the evaluation of data sets in semantics, mathematics, reasoning, code, and knowledge, GLM-4 has shown superior performance beyond Llama-3. In addition to multi-round conversations, GLM-4-Chat also has advanced features such as web browsing, code execution, custom tool calls (Function Call), and long text reasoning (supporting up to 128K context). This generation of models has added multi-language support, supporting 26 languages including Japanese, Korean, and German.
Model Intention: It is the latest generation of pre-trained models in the GLM-4 series launched by Zhipu AI
Model Info URL: https://huggingface.co/THUDM
Model License: License Info
Model Description: GLM-4 is the latest generation of pre-trained models in the GLM-4 series launched by Zhipu AI. In the evaluation of data sets in semantics, mathematics, reasoning, code, and knowledge, GLM-4 has shown superior performance beyond Llama-3. In addition to multi-round conversations, GLM-4-Chat also has advanced features such as web browsing, code execution, custom tool calls (Function Call), and long text reasoning (supporting up to 128K context). This generation of models has added multi-language support, supporting 26 languages including Japanese, Korean, and German.
Developer: https://huggingface.co/THUDM
File Size: 2627 MB
Context Length: 1024 tokens
Prompt Format:
{% for item in messages %}{% if item['role'] == 'system' %}<|system|>
{{ item['content'] }}{% elif item['role'] == 'user' %}<|user|>
{{ item['content'] }}{% elif item['role'] == 'assistant' %}<|assistant|>
{{ item['content'] }}{% endif %}{% endfor %}{% if add_generation_prompt %}<|assistant|>
{% endif %}
Template Name: glm
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Gemma 3n E2B it
Gemma 3n models are designed for efficient execution on low-resource devices. They are capable of multimodal input, handling text, image, video, and audio input, and generating text outputs, with open weights for pre-trained and instruction-tuned variants. These models were trained with data in over 140 spoken languages.
Model Intention: Gemma 3n models are designed for efficient execution on low-resource devices. They are capable of multimodal input
Model Info URL: https://huggingface.co/google/gemma-3n-E2B-it
Model License: License Info
Model Description: Gemma 3n models are designed for efficient execution on low-resource devices. They are capable of multimodal input, handling text, image, video, and audio input, and generating text outputs, with open weights for pre-trained and instruction-tuned variants. These models were trained with data in over 140 spoken languages.
Developer: https://huggingface.co/google
Update Date: 2025-06-27
File Size: 2720 MB
Context Length: 8000 tokens
Prompt Format:
Template Name: chatml
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
SmolLM2 1.7B
SmolLM2 was trained on 11 trillion tokens and demonstrates significant advances over other small models, particularly in instruction following, knowledge, reasoning, and mathematics.
Model Intention: SmolLM2 is capable of solving a wide range of tasks while being lightweight enough to run on-device
Model Info URL: https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B-Instruct
Model License: License Info
Model Description: SmolLM2 was trained on 11 trillion tokens and demonstrates significant advances over other small models, particularly in instruction following, knowledge, reasoning, and mathematics.
Developer: https://huggingface.co/HuggingFaceTB
Update Date: 2024-11-02
File Size: 1060 MB
Context Length: 8192 tokens
Prompt Format:
<|im_start|>system
{{system}}<|im_end|>
<|im_start|>user
{{prompt}}<|im_end|>
<|im_start|>assistant
Template Name: chatml
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Phi4 mini 4B
Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data. The model is intended for broad multilingual commercial and research use. The model provides uses for general purpose AI systems and applications which require: 1). Memory/compute constrained environments; 2). Latency bound scenarios; 3) Strong reasoning (especially math and logic). The model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
Model Intention: Phi-4-mini-instruct is a lightweight model focused on high-quality, reasoning dense data. It supports 128K token context length
Model Info URL: https://huggingface.co/microsoft/Phi-4-mini-instruct
Model License: License Info
Model Description: Phi-4-mini-instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data. The model is intended for broad multilingual commercial and research use. The model provides uses for general purpose AI systems and applications which require: 1). Memory/compute constrained environments; 2). Latency bound scenarios; 3) Strong reasoning (especially math and logic). The model is designed to accelerate research on language and multimodal models, for use as a building block for generative AI powered features.
Developer: https://huggingface.co/microsoft
File Size: 2020 MB
Context Length: 2048 tokens
Prompt Format:
{% for message in messages %}{% if message['role'] == 'system' and 'tools' in message and message['tools'] is not none %}{{ '<|' + message['role'] + '|>' + message['content'] + '<|tool|>' + message['tools'] + '<|/tool|>' + '<|end|>' }}{% else %}{{ '<|' + message['role'] + '|>' + message['content'] + '<|end|>' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>' }}{% else %}{{ eos_token }}{% endif %}
Template Name: llama3.2
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Qwen3 1.7B
Qwen3 1.7B is one of the small models in the Qwen series, designed for efficiency and speed. It can run seamlessly on edge devices, enabling rapid inference and real-time applications. This compact model is ideal for testing scenarios, prototyping, or deployment in resource-constrained environments.
Model Intention: The 1.7B model in the Qwen3 series is a small model designed for fast predictions and function calls.
Model Info URL: https://huggingface.co/Qwen/Qwen3-1.7B
Model License: License Info
Model Description: Qwen3 1.7B is one of the small models in the Qwen series, designed for efficiency and speed. It can run seamlessly on edge devices, enabling rapid inference and real-time applications. This compact model is ideal for testing scenarios, prototyping, or deployment in resource-constrained environments.
Developer: https://huggingface.co/Qwen
File Size: 1110 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
ERNIE-4.5 0.3B
ERNIE 4.5 is a series of open source models created by Baidu. The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations: 1. Multimodal Heterogeneous MoE Pre-Training; 2. Scaling-Efficient Infrastructure; 3. Modality-Specific Post-Training
Model Intention: ERNIE-4.5-0.3B-Base is a text dense Base model for testing the model's architecture.
Model Info URL: https://huggingface.co/baidu/ERNIE-4.5-0.3B-Base-PT
Model License: License Info
Model Description: ERNIE 4.5 is a series of open source models created by Baidu. The advanced capabilities of the ERNIE 4.5 models, particularly the MoE-based A47B and A3B series, are underpinned by several key technical innovations: 1. Multimodal Heterogeneous MoE Pre-Training; 2. Scaling-Efficient Infrastructure; 3. Modality-Specific Post-Training
Developer: https://huggingface.co/baidu
File Size: 233 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
SmolLM3 3B
SmolLM3 is a fully open model that offers strong performance at the 3B–4B scale. The model is a decoder-only transformer using GQA and NoPE (with 3:1 ratio), it was pretrained on 11.2T tokens with a staged curriculum of web, code, math and reasoning data. Post-training included midtraining on 140B reasoning tokens.
Model Intention: SmolLM3 is a 3B parameter language model designed to push the boundaries of small models. It supports 6 languages (English, French, Spanish, German, Italian, and Portuguese), advanced reasoning and long context.
Model Info URL: https://huggingface.co/HuggingFaceTB/SmolLM3-3B
Model License: License Info
Model Description: SmolLM3 is a fully open model that offers strong performance at the 3B–4B scale. The model is a decoder-only transformer using GQA and NoPE (with 3:1 ratio), it was pretrained on 11.2T tokens with a staged curriculum of web, code, math and reasoning data. Post-training included midtraining on 140B reasoning tokens.
Developer: https://huggingface.co/HuggingFaceTB
File Size: 1920 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
LFM2 1.2B
LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency. LFM2 is a new hybrid Liquid model with multiplicative gates and short convolutions. It supported languages: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
Model Intention: LFM2 1.2B is particularly suited for agentic tasks, data extraction, RAG, creative writing, and multi-turn conversations
Model Info URL: https://huggingface.co/LiquidAI/LFM2-1.2B
Model License: License Info
Model Description: LFM2 is a new generation of hybrid models developed by Liquid AI, specifically designed for edge AI and on-device deployment. It sets a new standard in terms of quality, speed, and memory efficiency. LFM2 is a new hybrid Liquid model with multiplicative gates and short convolutions. It supported languages: English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
Developer: https://huggingface.co/LiquidAI
File Size: 696 MB
Context Length: 1024 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Menlo Lucy 1.7B
Lucy is a compact but capable 1.7B model focused on agentic web search and lightweight browsing. It is built on Qwen3-1.7B and optimized to run efficiently on mobile devices, even with CPU-only configurations. It was developed by Alan Dao, Bach Vu Dinh, Alex Nguyen, and Norapat Buppodom.
Model Intention: Lucy is a compact but capable 1.7B model focused on agentic web search and lightweight browsing.
Model URL: https://huggingface.co/flyingfishinwater/good_and_small_models/resolve/main/Menlo_Lucy-Q4_K_M.gguf
Model Info URL: https://huggingface.co/Menlo/Lucy
Model License: License Info
Model Description: Lucy is a compact but capable 1.7B model focused on agentic web search and lightweight browsing. It is built on Qwen3-1.7B and optimized to run efficiently on mobile devices, even with CPU-only configurations. It was developed by Alan Dao, Bach Vu Dinh, Alex Nguyen, and Norapat Buppodom.
Developer: https://huggingface.co/Menlo
File Size: 1056 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Nemotron 1.5B
OpenReasoning-Nemotron-1.5B is a large language model (LLM) which is a derivative of Qwen2.5-1.5B-Instruct. It is a reasoning model that is post-trained for reasoning about math, code and science solution generation. This model is ready for commercial/non-commercial research use.
Model Intention: It is a reasoning model that is post-trained for reasoning about math, code and science solution generation.
Model Info URL: https://huggingface.co/nvidia/OpenReasoning-Nemotron-1.5B
Model License: License Info
Model Description: OpenReasoning-Nemotron-1.5B is a large language model (LLM) which is a derivative of Qwen2.5-1.5B-Instruct. It is a reasoning model that is post-trained for reasoning about math, code and science solution generation. This model is ready for commercial/non-commercial research use.
Developer: https://huggingface.co/nvidia
File Size: 940 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes
Qwen3 1.7B Uncensored
Qwen3 1.7B Uncensored is an unrestricted variant designed for creative writing and storytelling without content limitations. It excels at generating fiction stories, horror narratives, plot development, scene continuation, and roleplaying scenarios. This model provides unfiltered responses and can produce intense or graphic content, making it suitable for users seeking unrestricted AI interactions for creative purposes.
Model Intention: An uncensored 1.7B model optimized for creative writing, fiction stories, horror narratives, and unrestricted conversational scenarios.
Model Info URL: https://huggingface.co/DavidAU/Qwen3-1.7B-HORROR-Imatrix-Max-GGUF
Model License: License Info
Model Description: Qwen3 1.7B Uncensored is an unrestricted variant designed for creative writing and storytelling without content limitations. It excels at generating fiction stories, horror narratives, plot development, scene continuation, and roleplaying scenarios. This model provides unfiltered responses and can produce intense or graphic content, making it suitable for users seeking unrestricted AI interactions for creative purposes.
Developer: https://huggingface.co/DavidAU
File Size: 1110 MB
Context Length: 2048 tokens
Prompt Format:
Template Name: qwen
Add BOS Token: Yes
Add EOS Token: No
Parse Special Tokens: Yes