Best local AI models for NVIDIA GTX 760M

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 GTX 760M is a mobile graphics card equipped with 2 GB of GDDR5 memory. This physical memory size is the absolute limit for running artificial intelligence models directly on your hardware. To fit models within this tight limit you must use quantized versions. Quantization reduces the precision of model weights to save space. The quant column shows the best balance of quality and size for each model.

For models up to 2.8B parameters you can run entirely within your 2 GB of GDDR5 memory. The Allegro 2.8B model fits at the Q4_K_M quant which uses exactly 2 GB of memory. The Open-Sora Plan 2.7B model also uses 2 GB of memory at the Q4_K_M quant. Smaller models like the LFM2 2.6B and Playground v2.5 2.6B require 1.9 GB of memory at the Q4_K_M quant.

As model sizes decrease you can use higher quality quantization levels. Stable Diffusion 3.5 Medium 2.5B and Canary 2.5B use 1.8 GB of memory at the Q4_K_M quant. SeamlessM4T v2 2.3B and Parler-TTS 2.2B run at the Q5_K_M quant using 2 GB and 1.9 GB of memory. Models like SmolVLM 2B and Stable Diffusion 3 Medium 2B can use the Q6_K quant which fits exactly into 2 GB of memory.

Very small models can run at the highest quality Q8_0 quant. StableLM 2 1.6B and Sana 1.6B use 2 GB of memory at Q8_0. Whisper Large v3 1.55B also uses 2 GB of memory at Q8_0. ControlNet 1.5B and AudioGen 1.5B require 1.9 GB of memory at Q8_0. TinyLlama 1.1B runs at Q8_0 and uses only 1.4 GB of memory which leaves some safety margin.

When a model is too large for your 2 GB of GDDR5 memory you must use CPU offload. This process splits the model between your graphics card and your system RAM. CPU offload allows you to run larger models but it significantly reduces processing speed. We assume your computer has 32 GB of system RAM to handle these larger files.

With CPU offload you can run SmolLM3 3B or Replit Code v1.5 3B which need 2.2 GB of graphics memory at Q4_K_M and 4.2 GB of system RAM. MusicGen 3.3B needs 2.4 GB of graphics memory at Q4_K_M and 4.4 GB of system RAM. Stable Diffusion XL 3.417B needs 4.1 GB of graphics memory at FP8 and 6.1 GB of system RAM. SDXL Turbo 3.5B needs 2.6 GB of graphics memory at Q4_K_M and 4.6 GB of system RAM.

You must remember the 4k context caveat when running local text models. Generating longer responses or reading long prompts increases memory usage. The memory figures listed here are for basic generation. Running a model at a full 4k context window will require more memory than these base figures and might cause your system to run out of memory.