Yes — Phi-5 Large 28B (28B) runs at 26 tok/s on M3 Max with 64 GB RAM using Q4_K_M quantization via Ollama. First token latency is 1.1s. A capable open-source LLM with 28B parameters.
LLMCheck measured Phi-5 Large 28B on M3 Max using the standard methodology: Q4_K_M quantization, 256-token input, 512-token output, 3 runs averaged on a freshly-booted system.
| Metric | Value |
|---|---|
| Tokens per second | 26 tok/s |
| Time to first token | 1.1s |
| Quantization | Q4_K_M |
| Minimum RAM | 64 GB |
| Recommended engine | Ollama |
| Parameters | 28B |
| Benchmark date | 2026-07 |
Q4_K_M 28B Ollama M3 Max
The recommended engine for Phi-5 Large 28B on M3 Max is Ollama. Install Ollama, then pull the model:
Ollama handles quantization automatically — it will download the Q4_K_M variant (~64 GB) and start an interactive chat session.
| Chip | Speed | First Token | Min RAM | Engine |
|---|---|---|---|---|
| M5 Max | 38 tok/s | 0.6s | 128 GB | MLX |
| M5 Max | 34 tok/s | 0.7s | 64 GB | Ollama |
| M4 Pro | 28 tok/s | 0.9s | 32 GB | MLX |
To run Phi-5 Large 28B on M3 Max you need:
Phi-5 Large 28B needs about 64 GB of unified memory. These current Apple Silicon Macs have the headroom to run it comfortably:
Not sure which Mac fits your budget? See the best Mac for running this →
As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep our benchmarks free.
See how Phi-5 Large 28B stacks up against other models on your specific Mac hardware.
Open Compare Tool Full Leaderboard