My Mac
What can this Mac realistically run?
Enter what your Mac reports and the work you want to test. Get a cautious model range, real model examples to evaluate, compatible runtime paths, and the limits that still need measurement.
Cautious model range
7–9B class at Q4
A practical small-model range. Long context and other applications reduce headroom.
- Build
- 16 GB unified memory
- Work
- Verify every factual statement and remove unsupported additions.
- Evidence
- Memory-fit estimate; exact artifact not tested here.
Named examples
Models inside this cautious size range
These are publisher-verified model identities, not a quality ranking. You still need a compatible quantized artifact; license terms, artifact size, context, and runtime matter.
Runtime paths
Three ways to run the artifact
What this answer does not know
- Tokens per second on your exact chip and runtime.
- Whether a named model is good at your exact task.
- How much context, cache, or concurrency your workflow needs.
- Whether the complete system is private, secure, or compliant.
Read the answer correctly
Memory says what may fit. Testing says what is useful.
The chip generation and tier can materially change speed, but Ground Floor will not invent a speed number without the exact build, artifact, runtime, context, settings, and retained run.
This explorer provides
- A conservative Q4 size class from unified memory.
- Real model identities inside that class.
- Runtime paths with primary-source links.
A direct test must provide
- Actual speed, memory pressure, and useful context.
- Task errors, omissions, edits, and review time.
- The complete local data path and controls.
Sources checked July 19, 2026: Apple M5 Pro/Max specifications, publisher model cards linked above, MLX-LM, llama.cpp, and LM Studio requirements.