Hugging Face × ModelScope
Model Atlas

Model Atlas

Astra integrates with both major hubs — search and install in one tap. Below is a curated set; the full catalog lives in the app.

Built-in
Qwen
Qwen3.5-0.8B-Instruct
0.8B

Alibaba's lightweight Tongyi Qianwen instruct model — Astra's built-in default, excellent in Chinese.

12.3kGGUF Q4_K_M
ModelScope
Qwen
Qwen2.5-1.5B-Instruct
1.5B

A stronger bilingual chat model with great long-context performance.

8.7kGGUF Q4
ModelScope
Meta
Llama-3.2-1B
1B

Meta's newest ultra-light Llama — extremely fast.

24.5kGGUF Q5_K_M
Hugging Face
Meta
Llama-3.2-3B
3B

Llama 3.2 mid-size — strong multilingual abilities.

31.2kGGUF Q4_K_M
Hugging Face
Google
Gemma-2-2B-it
2B

Google Gemma 2 instruct — balanced quality and efficiency.

15.8kGGUF Q5
Hugging Face
DeepSeek
DeepSeek-R1-Distill-1.5B
1.5B

DeepSeek R1 distilled — strong chain-of-thought reasoning.

18.4kGGUF Q4
ModelScope
DeepSeek
DeepSeek-Coder-1.3B
1.3B

Specialized in code completion and generation across languages.

9.6kGGUF Q4
Hugging Face
Microsoft
Phi-3.5-mini-instruct
3.8B

Microsoft's tiny powerhouse — exceptional parameter efficiency.

21.3kGGUF Q4
Hugging Face
01.AI
Yi-1.5-6B-Chat
6B

01.AI's bilingual Chinese-English chat model.

7.2kGGUF Q3_K_M
ModelScope
Mistral
Ministral-3B
3B

Mistral's on-device flagship — excellent performance.

11.5kGGUF Q4
Hugging Face
Qwen
Qwen2-VL-2B
2B

Multimodal model with image understanding.

6.8kGGUF Q4
ModelScope
Stability
StableLM-Zephyr-3B
3B

Stability AI's on-device chat model.

5.4kGGUF Q4
Hugging Face