The AI Bench
96 GB · 1,200 GB/s

What can the Mac Studio (M5 Ultra, 96 GB) run?

The most capable Mac for local AI. Runs 70–120B models.

Best models for each task

Our top three for this machine. Chat and image tasks assume an 8K-token conversation; coding, documents and agents assume 32K.

Chat & writing23 of 25 run

  1. Qwen3.8 Flash Next 125B-A6B (+51B n-gram emb.)Just fitsUD-IQ3_XXS · needs 84.1 GB at 8K · 50–110tok/sFast
  2. Qwen3.8 27BRuns greatQ8_0 · needs 31.6 GB at 8K · 18–30tok/sSmooth
  3. Gemma 4 31BRuns greatQ8_0 · needs 35.7 GB at 8K · 16–25tok/sSmooth

Coding23 of 26 run

  1. Qwen3.8 Flash Next 125B-A6B (+51B n-gram emb.)Just fitsUD-IQ3_XXS · needs 84.7 GB at 32K · 50–100tok/sFast
  2. Qwen3.8 27BRuns greatQ8_0 · needs 33.2 GB at 32K · 17–30tok/sSmooth
  3. Muse Glimmer 30BRuns greatQ8_0 · needs 32.5 GB at 32K · 18–30tok/sSmooth

Long documents17 of 20 run

  1. Qwen3.8 Flash Next 125B-A6B (+51B n-gram emb.)Just fitsUD-IQ3_XXS · needs 84.7 GB at 32K · 50–100tok/sFast
  2. Qwen3.8 27BRuns greatQ8_0 · needs 33.2 GB at 32K · 17–30tok/sSmooth
  3. Qwen3.6 35B-A3BRuns greatQ8_0 · needs 40.4 GB at 32K · 45–100tok/sFast

Agents & tools23 of 26 run

  1. Qwen3.8 Flash Next 125B-A6B (+51B n-gram emb.)Just fitsUD-IQ3_XXS · needs 84.7 GB at 32K · 50–100tok/sFast
  2. Qwen3.8 27BRuns greatQ8_0 · needs 33.2 GB at 32K · 17–30tok/sSmooth
  3. Qwen3.6 35B-A3BRuns greatQ8_0 · needs 40.4 GB at 32K · 45–100tok/sFast

Images in12 of 13 run

  1. Qwen3.8 Flash Next 125B-A6B (+51B n-gram emb.)Just fitsUD-IQ3_XXS · needs 84.1 GB at 8K · 50–110tok/sFast
  2. Qwen3.8 27BRuns greatQ8_0 · needs 31.6 GB at 8K · 18–30tok/sSmooth
  3. Qwen3.6 35B-A3BRuns greatQ8_0 · needs 39.9 GB at 8K · 50–110tok/sFast

Make images6 of 6 run

  1. Stable Diffusion XL Base 1.0Runs greatFP16 · needs 7.9 GB
  2. FLUX.2 [klein] 4BRuns greatBF16 · needs 14.0 GB
  3. Z-Image-TurboRuns greatBF16 · needs 17.0 GB

Speak text7 of 7 run

  1. Qwen3-TTS 1.7BRuns greatBF16 · needs 5.5 GB
  2. Kokoro-82MRuns greatFP32 · needs 1.4 GB
  3. Chatterbox Multilingual V3Runs greatFull · needs 4.2 GB

Transcribe7 of 7 run

  1. Voxtral Mini 4B RealtimeRuns greatBF16 · needs 17.0 GB
  2. Whisper large-v3-turbo (via faster-whisper)Runs greatFP16 · needs 2.6 GB
  3. Nemotron 3.5 ASR Streaming 0.6BRuns greatFP32 · needs 3.5 GB

Biggest model it runs well

Qwen3.5 122B-A10B (Q4_K_M, 76.5 GB file) at about 35–75tok/sFast.

Should you buy it?

Buy it if

You want the biggest models in a quiet box.

Skip it if

Your models fit in 64 GB — the M5 Max Studio is much cheaper.

Watch out for

  • Ultra chips reach less of their bandwidth than the numbers suggest; our estimate accounts for this.