Best local AI models for AMD R7 M365X

The largest models that fit fully

The 28 largest of the 28 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 M365X is an entry level mobile graphics card equipped with 1 GB of GDDR5 video memory. This hardware memory size is the absolute limit for running local AI models entirely on your GPU. To load and run a model successfully without system slowdowns, the model files and active processing data must fit within this 1 GB boundary.

To make models fit this compact memory space, developers use quantization. The quant column shows the specific compression level needed to run each model. For example, Tango 2 at 1.4B parameters requires a Q4_K_M quantization to fit exactly 1 GB of video memory. Smaller models like Qwen3 0.6B can run at Q8_0 quantization using 0.8 GB of video memory, while very small models like StyleTTS 2 at 0.15B can run uncompressed at FP16 using 0.4 GB of video memory.

When a model exceeds the local 1 GB video memory, you must use CPU offload. This method splits the workload between your GPU and your system RAM. We assume your computer has 32 GB of system RAM for these scenarios. For example, running Stable Diffusion 1.5 requires 1.3 GB of video memory at FP8 and also needs 3.3 GB of system RAM. CPU offload allows you to run larger tools like Whisper Large v3 or StableLM 2 1.6B, but processing speeds will be much slower than running fully on the GPU.

Running text models on this hardware comes with a strict context limit caveat. The memory figures listed are calculated for short interactions. If you increase the context window to 4k tokens, the active memory usage will grow quickly. This extra data can easily overflow your 1 GB video memory and force the system to slow down.

By selecting the correct quantization level, you can run diverse local AI tasks on your AMD R7 M365X. You can run text generation with TinyLlama 1.1B at Q5_K_M or Llama 3.2 1B at Q6_K. For audio and speech, you can use Bark at Q6_K or F5-TTS at FP16. Image generation is also possible using PixArt-α at Q8_0 or Stable Diffusion 2.1 at Q6_K.