Apple Mac Studio (M2 Ultra 192GB)

Scored against every model in our directory at Q4_K_M, using the ~144GB of GPU-usable memory on this machine and its 800 GB/s of memory bandwidth. Of 30 models, 28 run comfortably, 0 are a tight fit, and 2 will not load. The largest model that fits well here is OpenAI gpt-oss-120b.

How these are estimated: tokens/sec ≈ memory bandwidth × 0.65 ÷ active weights in GB. These are arithmetic estimates from published specifications, not lab measurements — BFWR does not run a hardware test lab. Real throughput varies with runner, quantization, context length and thermals.

Diffusion models size up differently from language models: the text encoder is often as large as the image model itself, and memory scales with resolution. Measured at 1024×1024, one image at a time, against the ~144GB of GPU-usable memory here, 11 of 11 open image models will run — the best being FLUX.2 [dev] at BF16 / FP16, peaking near 115.6GB — but that model is non-commercial; Qwen-Image is the best option you can ship.

Weight sizes are read from the real .safetensors files in each model’s Hugging Face repository. Activation memory is an approximation that varies with the attention implementation and sampler, so leave headroom. These are planning figures, not benchmarks.

What is the largest model the Apple Mac Studio (M2 Ultra 192GB) can run?

OpenAI gpt-oss-120b (116.8B (5.1B active)) is the largest model that fits, at roughly 65GB in Q4_K_M against ~144GB of GPU-usable memory. It is rated "Comfortable" here, with an estimated 183.2 tokens/sec.

How many tokens per second should I expect?

Speed depends on the model, not just the machine. On this hardware the estimates range from about 288.9 tokens/sec for the smallest models down to single digits for the largest that still fit. The estimate uses memory bandwidth (800 GB/s here) × 0.65 ÷ active weights in GB, and is not a measured benchmark.

Can the Apple Mac Studio (M2 Ultra 192GB) run a 70B model?

Yes — OpenAI gpt-oss-120b is rated "Comfortable" here at roughly 65GB in Q4_K_M, with an estimated 183.2 tokens/sec. Expect to keep the context modest.

How much context can I use?

128k tokens is a sensible ceiling on this machine at FP16 / Q8_0. Context is stored in the KV cache, which grows with every token and competes with the model weights for the same memory, so a longer context means a smaller model.

Can the Apple Mac Studio (M2 Ultra 192GB) generate images with Flux or Stable Diffusion?

Yes. 11 of 11 open image models fit at 1024x1024, the best being FLUX.2 [dev] at BF16 / FP16, peaking near 115.6GB. Note that FLUX.2 [dev] is licensed for non-commercial use only; Qwen-Image is the best option you can actually ship. Higher resolutions and larger batches cost proportionally more.

Which runner works best on the Apple Mac Studio (M2 Ultra 192GB)?

Ollama, LM Studio, AnythingLLM are all expected to run on this hardware. Ollama is the simplest starting point, and `ollama run gpt-oss:120b` is a good first model for this machine.

In-depth reads that pair with this machine — sizing memory, picking a runner, and choosing quantization for ~144GB of usable memory.

• Best Mac for Local LLMsMac mini M6 and M5 Pro vs Mac Studio M5 Max and Ultra.Read the guide
• Apple Silicon vs Snapdragon X for Local AIUnified memory, NPUs and what actually matters.Read the guide
• How to Run Qwen3.8-Flash-Next Locally180B with 6B active: which build fits 128GB, real speeds, licence.Read the guide