I have been curious about this: how does the new little Mac Mini M6 actually compare with my much larger, but older Mac Studio running an M1 Max?
The obvious assumption is that the Mac Studio should win. It has the Max chip, more memory bandwidth and a much bigger chassis. But the M6 is a much newer generation, so I wanted to run some real tests rather than rely on assumptions.
The short version? The Mini is an absolute little powerhouse for CPU work. The Studio still has some very important advantages, especially for local LLM generation.
The machines
CPU and system benchmarks
The CPU result was the first surprise. The M6 was not just a little quicker, it was dramatically quicker in both single-core and multi-core testing.
The M1 Max still delivered the higher memory bandwidth, which is exactly what I expected from its wider memory system. That distinction becomes important when running local AI models.
The Qwen LLM test
The small model running on the Mini was Qwen3.5:2b. I installed the same Ollama tag on the Studio and tested generation speed, prompt processing and two simultaneous requests.
There is an important qualification here: Ollama resolved the tag to platform-specific variants rather than bit-identical model files.
So these are practical results for the installed configurations, not a perfectly controlled model-architecture comparison.
Qwen generation speed
This is where the old Studio fights back. The M1 Max was substantially faster at actually generating answer tokens. If the workload is a local chatbot that is continuously producing text, the Studio feels much more capable.
Prompt processing
This was the other interesting result. The Mini was faster at ingesting prompts, even though the Studio was faster at generating the response. That could matter for document analysis, large context windows and workloads where a lot of input is being pushed through the model.
What does this actually mean?
For normal computing, compiling, scripting, automation and CPU-heavy work, the M6 Mac Mini is the clear winner in these tests. It is genuinely impressive how much performance is packed into such a small and quiet machine.
For local LLM use, the answer is more nuanced:
My overall take is that the M6 Mac Mini really is a little powerhouse, just not in exactly the same way as the M1 Max Mac Studio. The Mini has the newer, faster CPU. The Studio still has the wider memory system and the stronger local LLM generation performance.
The benchmark code and the full test notes are available if anyone wants to reproduce or challenge the results.
The obvious assumption is that the Mac Studio should win. It has the Max chip, more memory bandwidth and a much bigger chassis. But the M6 is a much newer generation, so I wanted to run some real tests rather than rely on assumptions.
The short version? The Mini is an absolute little powerhouse for CPU work. The Studio still has some very important advantages, especially for local LLM generation.
The machines
- Mac Mini M6: 12 logical CPUs and 16 GB unified memory
- Mac Studio M1 Max: 10 logical CPUs
- The tests ran locally on each Mac. I connected to them remotely over Tailscale.
CPU and system benchmarks
| Test | Mac Mini M6 | Mac Studio M1 Max | Winner |
|---|---|---|---|
| CPU single-core | 396.1M rounds/sec | 260.5M rounds/sec | M6 +52% |
| CPU multi-core | 4,052.5M rounds/sec | 2,322.8M rounds/sec | M6 +74% |
| Memory bandwidth | 124.9 GB/sec | 168.6 GB/sec | M1 Max +35% |
| SHA-256 hashing | 3.53 GB/sec | 2.38 GB/sec | M6 +49% |
| gzip -1 compression | 0.36 sec | 0.61 sec | M6 +70% |
| Synchronous 1 GB write | 4.4 GB/sec | 1.5 GB/sec | M6 about 3x |
The CPU result was the first surprise. The M6 was not just a little quicker, it was dramatically quicker in both single-core and multi-core testing.
The M1 Max still delivered the higher memory bandwidth, which is exactly what I expected from its wider memory system. That distinction becomes important when running local AI models.
The Qwen LLM test
The small model running on the Mini was Qwen3.5:2b. I installed the same Ollama tag on the Studio and tested generation speed, prompt processing and two simultaneous requests.
There is an important qualification here: Ollama resolved the tag to platform-specific variants rather than bit-identical model files.
- Mac Mini: Qwen3.5, 2.2B parameters, MXFP8, MLX runner
- Mac Studio: Qwen3.5, 1.9B parameters, Q8_0, GPU runner
So these are practical results for the installed configurations, not a perfectly controlled model-architecture comparison.
Qwen generation speed
| Test | Mac Mini M6 | Mac Studio M1 Max | Result |
|---|---|---|---|
| 128-token generation | 55.0 tokens/sec | 101.1 tokens/sec | M1 Max +84% |
| Two simultaneous requests | 53.9 aggregate tokens/sec | 97.0 aggregate tokens/sec | M1 Max +80% |
| Cold model load | 0.43 sec | 1.33 sec | M6 faster |
This is where the old Studio fights back. The M1 Max was substantially faster at actually generating answer tokens. If the workload is a local chatbot that is continuously producing text, the Studio feels much more capable.
Prompt processing
| Prompt size | Mac Mini M6 | Mac Studio M1 Max | Result |
|---|---|---|---|
| About 1,054 prompt tokens | 2,349 tok/sec | 1,573 tok/sec | M6 +49% |
| Longer prompt | About 2x the Studio rate | Baseline | M6 ahead |
This was the other interesting result. The Mini was faster at ingesting prompts, even though the Studio was faster at generating the response. That could matter for document analysis, large context windows and workloads where a lot of input is being pushed through the model.
What does this actually mean?
For normal computing, compiling, scripting, automation and CPU-heavy work, the M6 Mac Mini is the clear winner in these tests. It is genuinely impressive how much performance is packed into such a small and quiet machine.
For local LLM use, the answer is more nuanced:
- The M6 is excellent at prompt ingestion and cold-start time.
- The M1 Max is much faster at generating tokens.
- The M1 Max also has higher memory bandwidth, which helps explain its advantage with local AI workloads.
My overall take is that the M6 Mac Mini really is a little powerhouse, just not in exactly the same way as the M1 Max Mac Studio. The Mini has the newer, faster CPU. The Studio still has the wider memory system and the stronger local LLM generation performance.
The benchmark code and the full test notes are available if anyone wants to reproduce or challenge the results.
Last edited: