Best local AI models for Intel Arc Pro A60

The largest models that fit fully

The 30 largest of the 147 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 Intel Arc Pro A60 workstation graphics card features 12 GB of GDDR6 dedicated video memory. This memory capacity determines which artificial intelligence models can run entirely on the hardware. When a model fits completely within this VRAM limit, it achieves the fastest processing speeds. If a model exceeds this limit, some data must spill over into the system memory, which slows down performance.

To fit larger models into the 12 GB limit, developers use quantization. Quantization reduces the precision of the model weights to save space. The best quant column shows the optimal balance between model size and accuracy. For example, the DeepSeek-Coder-V2 16B model fits in 11.7 GB of VRAM using the Q4_K_M quantization level. Similarly, the Phi-4 14B model uses the Q5_K_M quantization level to fit within 11.9 GB of VRAM.

For models that require slightly more than 12 GB of memory, you can use CPU offloading. This technique shares the workload between your graphics card and your system RAM. If you have 32 GB of system RAM, you can run larger models with a minor speed penalty. Under this setup, the FLUX.1 dev 12B model requires 14.4 GB of memory at FP8 or optimized settings, which uses 16.4 GB of system RAM. The Codestral 22B model requires 16.1 GB of memory at the Q4_K_M quantization level, which utilizes 18.1 GB of system RAM.

Several highly capable models fit directly onto the Intel Arc Pro A60 without offloading. The Gemma 3 12B and Mistral NeMo 12B models both run at the Q6_K quantization level using 11.8 GB of VRAM. For image generation, the FLUX.1 schnell 12B model also runs at the Q6_K quantization level using 11.8 GB of VRAM. Smaller models like the GLM-4 9B and Yi-Coder 9B can run at the higher precision Q8_0 quantization level using 11.4 GB of VRAM.

When planning your local deployments, you must account for the context window. The memory figures listed here assume a standard 4k context window. If you increase the context window to process longer documents or larger chat histories, the memory usage will increase. This extra memory requirement may force you to use a lower quantization level or rely on CPU offloading to prevent out of memory errors.