Best local AI models for NVIDIA MX250

The largest models that fit fully

The 30 largest of the 56 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 NVIDIA MX250 is an entry level laptop graphics card equipped with 2 GB of GDDR5 memory. This dedicated video memory determines which artificial intelligence models can run entirely on your graphics hardware. When a model fits completely inside this 2 GB limit, it processes tokens and images at maximum speed because the system does not need to transfer data across the slow system bus.

To fit modern models into this compact memory space, developers use quantization. The quant column shows the specific compression level required to run each model. For example, the Allegro 2.8B model fits inside 2 GB of video memory when compressed to a Q4_K_M quantization. Smaller models like Moondream 2 can run at a higher quality Q6_K quantization while using 1.9 GB of memory. TinyLlama 1.1B can run at an uncompressed Q8_0 quantization because it only requires 1.4 GB of memory.

If a model requires more than 2 GB of memory, you must use CPU offload. This technique splits the model between your graphics card and your system RAM. Running the 3B parameter SmolLM3 requires 2.2 GB of video memory at Q4_K_M quantization and an additional 4.2 GB of system RAM. While CPU offload allows you to run larger models like MusicGen or Stable Diffusion XL, it reduces processing speed because your system RAM is much slower than your graphics memory.

When running text models, you must also consider the context window. The memory figures listed here assume a standard 4k context window. If you input long documents or maintain long conversations, the memory usage will increase. This extra data can push a model that normally fits in 2 GB over the limit, which forces the system to use slow CPU offload.

For image generation, you can run Stable Diffusion 3 Medium at Q6_K quantization using exactly 2 GB of video memory. If you want to use larger architectures like Stable Diffusion XL, you will need to offload data. Stable Diffusion XL requires 4.1 GB at FP8 or optimized settings, which uses your full video memory and 6.1 GB of system RAM.