Memory at each precision
| Version | Weights | With context | Smallest Mac |
|---|---|---|---|
| 8-bit | 31.9 GB | 34.5 GB | 48 GB |
| 6-bit | 24.4 GB | 26.4 GB | 36 GB |
| 4-bit | 16.9 GB | 18.4 GB | 32 GB |
| 3-bit lower quality | 13.2 GB | 14.7 GB | 24 GB |
| 2-bit lower quality | 9.4 GB | 10.9 GB | 16 GB |
On every Apple chip
The smallest memory size of each chip that runs Hy-Memory-30B-A3B, the best version that fits it, and the writing speed you can expect.
| Chip | Needs | Version | Tokens/s |
|---|---|---|---|
| M6 | 32 GB | 4-bit Runs well | 43–76 |
| M5 Ultra | 96 GB | 8-bit Runs well | 159–282 |
| M5 Max | 36 GB | 6-bit Runs well | 80–142 |
| M5 Pro | 48 GB | 8-bit Runs well | 41–72 |
| M5 | 32 GB | 4-bit Runs well | 38–68 |
| M4 Max | 36 GB | 6-bit Runs well | 71–126 |
| M4 Pro | 48 GB | 8-bit Runs well | 36–64 |
| M4 | 32 GB | 4-bit Runs well | 30–53 |
| M3 Ultra | 96 GB | 8-bit Runs well | 108–193 |
| M3 Max | 36 GB | 6-bit Runs well | 52–92 |
| M3 Pro | 36 GB | 6-bit Runs well | 26–46 |
| M3 | 16 GB | 2-bit Runs well | 45–80 |
| M2 Ultra | 64 GB | 8-bit Runs well | 106–188 |
| M2 Max | 32 GB | 4-bit Runs well | 100–178 |
| M2 Pro | 32 GB | 4-bit Runs well | 50–89 |
| M2 | 16 GB | 2-bit Runs well | 45–80 |
| M1 Ultra | 64 GB | 8-bit Runs well | 106–188 |
| M1 Max | 32 GB | 4-bit Runs well | 100–178 |
| M1 Pro | 32 GB | 4-bit Runs well | 50–89 |
| M1 | 16 GB | 2-bit Runs well | 31–54 |
How to run it
- The easy way: download MLX Harness, free and open source. Search for Hy-Memory-30B-A3B, download the version that fits your Mac, and chat. It also serves the model to your other apps over an OpenAI-compatible API.Download MLX HarnessFree · Apple Silicon · macOS 14.5 or later
- In Terminal with Apple’s MLX:
pip install mlx-lmmlx_lm.chat --model mlx-community/Hy-Memory-30B-A3B-4bit --trust-remote-code--trust-remote-code lets mlx-lm run the Python code that ships with this build, which it needs to load.Build: mlx-community/Hy-Memory-30B-A3B-4bit on Hugging Face. Other precisions sit next to it.
Step by step: download and run LLMs on a Mac · how to use MLX Harness · MLX vs Ollama vs llama.cpp
About Hy-Memory-30B-A3B
| Made by | Tencent |
|---|---|
| Parameters | 30.1B total · 3.2B active (mixture of experts) |
| Context | 256K tokens |
| Licence | Apache-2.0 |
| Released | 2026-10 |
| Input | Text |
| Weights | tencent/Hy-Memory-30B-A3B |
Questions
- How much memory does Hy-Memory-30B-A3B need?
- About 16.9 GB of weights at 4-bit, plus room for the context: 18.4 GB in all. On a Mac that means 32 GB of unified memory or more, since macOS lets the GPU use about 75% of it by default.
- How fast is Hy-Memory-30B-A3B on a Mac?
- Writing speed depends on memory bandwidth. The fastest Mac for it here is an M5 Ultra with 96 GB at about 159–282 tokens a second (8-bit). The smallest Mac that runs it, with 32 GB, manages about 100–178 on an M2 Max.
- Why does Hy-Memory-30B-A3B run faster than its size suggests?
- It is a mixture-of-experts model: it holds 30.1B parameters in memory but reads only 3.2B for each token, so it writes at the speed of a much smaller model.
Similar models
For your model card or README
Shows the smallest Mac that runs Hy-Memory-30B-A3B well and links here.
[](https://mlx.app/run/hy-memory-30b-a3b)How it’s worked out
Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.
Open data: mlx.app/data (JSON, CSV)