Best local AI models for AMD R7 M380

The largest models that fit fully

The 30 largest of the 81 models that fit; every smaller model in the catalog fits too. Best quant means the highest quality compression whose weights and 4k context both sit inside the memory.

Close, but only with CPU offload

These need more than the card holds at their smallest practical quant, so part of the model runs from system memory (figures assume 32 GB of it). They work, several times slower.

How to read this

The AMD Radeon R7 M380 is an entry level laptop graphics card equipped with 4 GB of DDR3 video memory. This dedicated memory size determines the maximum size of the artificial intelligence models you can run entirely on the graphics hardware. Because DDR3 memory has lower bandwidth than modern GDDR memory types, keeping the entire model within the onboard VRAM is critical for maintaining usable processing speeds.

To fit models into this 4 GB limit, you must use quantized versions. Quantization is a compression method that reduces the precision of model weights. The best quant column shows the optimal balance of size and quality for this hardware. For example, a 5B model like Lumina-Next or CogVideoX 2B / 5B can run at the Q4_K_M quantization level using 3.7 GB of VRAM. Smaller models like SmolLM3 3B or Replit Code v1.5 3B can run at the higher quality Q8_0 quantization level using 3.8 GB of VRAM.

When a model exceeds the 4 GB VRAM limit, you must use CPU offload. This technique splits the model layers between your graphics card and your system RAM. Assuming your computer has 32 GB of system RAM, you can run larger models like Mistral 7B or Qwen2.5 7B. A Q4_K_M quantization of Mistral 7B needs 5.7 GB of memory, which requires 7.7 GB of system RAM to handle the offloaded layers. While CPU offload allows you to run these larger models, the transfer of data over the system bus will significantly slow down generation speeds.

You can also run specialized vision and image generation models on this hardware. DeepSeek-VL2 fits within 3.8 GB of VRAM using the Q5_K_M quantization. For image generation, Stable Diffusion 3.5 Medium fits within 3.2 GB of VRAM using the Q8_0 quantization. If you choose to run Stable Diffusion XL, the model needs 4.1 GB of memory at FP8 or optimized settings, which requires 6.1 GB of system RAM through CPU offload.

When running language models on a 4 GB card, you must consider the context window size. The standard 4k context window requires additional VRAM to store active conversation history. If your model usage approaches the 4 GB limit, a large context history can cause the system to run out of memory. You may need to limit your active context window or choose smaller models like Danube 3 4B or Phi-3.5 Mini to ensure stable operation.