M5 LLM benchmark
No signed M5 submissions yet. Check the configuration guidance below, or contribute a reproducible measurement from your own machine.
Specs vendor specs, not measured
- Unified memory (max configuration)
- 32 GB
- Memory bandwidth
- 153 GB/s
- Launch date
- 2025-10-15
Sources: support.apple.com, apple.com. Vendor-published figures, accessed 2026-10-14.
Plan memory before choosing a model
The catalog lists a family maximum of 32 GB of unified memory. Enter your installed amount; smaller configurations have less space. Model weights, context cache and runtime overhead must share the available memory. Capacity alone does not establish which model will fit or how fast it will run. Use the memory-fit calculator for your model and context as an estimate, then verify with the intended runtime. No signed M5 measurement is available yet.
Contribute a signed M5 benchmark
Run on YOUR hardware to populate this page: pipx install llm-speed && llm-speed bench
$ pipx install llm-speed && llm-speed bench
Community folklore on M5
3 unverified claims extracted from Reddit/HN comments. Lower trust than signed runs above; every row links to the source.
- communityconfidence 60%
72.00tok/s — Qwen3-Coder-Next on M5 via mlx
our signed data: M5 · Qwen3-Coder-Next
“128GB RAM - Qwen3 Coder Next 8-Bit Benchmark # Qwen3-Coder-Next 8-Bit Benchmark: MLX vs Ollama **TLDR**: M5-Max with 128gb of RAM gets 72 tokens per second from Qwen3-Coder-Next 8-Bit using MLX Overview This benchmark compares two local inference backends — **MLX** (Ap…”
- communityconfidence 55%
128.0tok/s — on M5 via llama.cpp FP16
our signed data: M5
“± 3.55| |llama 7B Q4\_0|3.56 GiB|6.74 B|Vulkan|100|1|tg128|22.31 ± 0.06| Intel: |Build|Hardware|Backend|FP16 TFLOPS|MBW GB/s|pp512 t/s|tg128 t/s|t/TFLOP|MBW %| |:-|:-|:-|:-|:-|:-|:-|:-|:-| |b4008|Arc 140V|IPEX-LLM|32.0|136.5|656.5|22.98|20.52|59.93| Admittedly. the Intel data …”
- communityconfidence 55%
9.00tok/s — on M5 via mlx Q4
our signed data: M5
“cBook Pro with ~ 6-9gb of RAM using MLX. https://github.com/danveloper/flash-moe It's because it's an MoE, you can offload to SSD. I get ~9t/s on my M5, danveloper was using an M3. It would probably run on my M1 but much more slowly because of SSD read bandwidth. Prompt proces…”
Request a run on M5
No measured run yet for these pairs. Vote for the one you want; the most-requested runs go first. See all requests.
We never sell or share your email. Privacy