Best local AI models for Apple M2 Ultra

The largest models that fit fully

The 30 largest of the 186 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.

How to read this

The Apple M2 Ultra chip with a 64 GB unified memory pool provides 51.2 GB of usable memory dedicated to running local AI models. This specific allocation is the maximum space available for both the model weights and the active context window. If a model exceeds this limit it will fail to run or experience extreme performance slowdowns. Keeping your model size within this 51.2 GB boundary ensures that execution remains entirely on the high speed unified memory of the chip.

To fit larger models into this memory space you must use quantized versions. The quantization level determines how much the original model weights are compressed. For example a Q8_0 quantization represents an 8 bit format that preserves high accuracy while reducing the footprint. A Q6_K quantization uses a 6 bit format to compress even larger models so they can fit inside the 51.2 GB limit.

The largest models that fit this system include Jamba 1.5 Mini / Large 52B and Llama 3.1 Nemotron 51B. Both of these models require a Q6_K quantization. Jamba 1.5 Mini / Large 52B uses exactly 51.2 GB of memory while Llama 3.1 Nemotron 51B uses 50.2 GB of memory. Mixtral 8x7B also fits at a Q6_K quantization using 46.2 GB of memory.

Many powerful models fit at the higher quality Q8_0 quantization. Seed-OSS 36B uses 45.8 GB of memory. Qwen3.6-35B-A3B and Command R (35B) both use 44.5 GB of memory. You can also run Yi 1.5 9B / 34B Granite Code 3B to 34B LLaVA 1.5 / 1.6 (7B to 34B) and Ovis 2 at Q8_0 quantization with each using 43.2 GB of memory.

Other excellent options at Q8_0 quantization include DeepSeek-Coder 1.3B / 6.7B / 33B and WizardCoder 33B which both use 42 GB of memory. OTel 2.0 LLM 31B IT uses 44.9 GB of memory. Qwen3 8B / 14B / 32B Qwen3.5 (dense variants) Aya Expanse 8B / 32B Granite 4.0 Small/Tiny and Qwen2.5-Coder 0.5B to 32B all run at Q8_0 quantization using 40.7 GB of memory.

Smaller models leave more room for processing. Qwen3-30B-A3B and Qwen3-Coder 30B-A3B use 38.2 GB of memory. Gemma 3 27B Gemma 3 4B/12B/27B (vision) and Wan 2.2 / 2.5 use 34.3 GB of memory. Gemma 4 26B-A4B and Gemma 4 (all sizes) use 33.1 GB of memory. Aria uses 31.8 GB of memory. Mistral Small 3.2 Magistral Small and Devstral Small 1.1 use 30.5 GB of memory. Solar Pro uses 28 GB of memory.

All memory calculations are based on a standard 4k context window. If you increase the context window to process longer documents the memory usage will rise. This extra memory requirement might force you to choose a lower quantization level to prevent exceeding the 51.2 GB limit. There are no CPU offload cases for this configuration because the system does not offload to system RAM.