Yes: Qwen 3.5 9B runs on 53 of 54 Macs in the LLMCheck index. It needs about 7 GB at a 4K context, 10 GB at 64K. The smallest and cheapest current Mac that runs it at 15+ tok/s is the Mac mini M6 16 GB ($899; 30 tok/s community-measured). Best measured figure: 167 tok/s on an M5 Ultra (community).
One row per Mac configuration, grouped by product line. The build is the one each speed is for; the context column is the longest prompt that fits inside macOS’s default GPU memory limit with a 16-bit KV cache (“raised” where the limit has to be raised first; “full” when the whole window fits). Speeds are decode tok/s from the LLMCheck index, each with its provenance mark. For dense models, the formula’s typical error is about 9% (the median gap to 75 measured plain-decode runs; it tends to undershoot (median −6%), and an estimate scaled from the model’s own measured run is usually closer; method).
| Mac | Verdict | Build | Speed | Max context | Model page |
|---|---|---|---|---|---|
| MacBook Air | |||||
| MacBook Air M5 16 GB | Runs | Q4_K_M | estimated27 tok/s | 148K | — |
| MacBook Air M5 24 GB | Runs | Q4_K_M | estimated27 tok/s | 256K (full) | — |
| MacBook Air M5 32 GB | Runs | Q4_K_M | estimated27 tok/s | 256K (full) | — |
| MacBook Air M1 8 GB refurb | Over GPU limit why ↓ | Q4_K_M | estimated10 tok/s | — | M1 page |
| MacBook Air M4 16 GB used | Runs | Q4_K_M | estimated18 tok/s | 113K | M4 page |
| MacBook Air M2 16 GB refurb | Runs | Q4_K_M | estimated15 tok/s | 113K | — |
| MacBook Air M3 16 GB refurb | Runs | Q4_K_M | estimated15 tok/s | 113K | M3 page |
| MacBook Air M1 16 GB refurb | Runs, slowly | Q4_K_M | estimated10 tok/s | 113K | M1 page |
| MacBook Air M4 24 GB used | Runs | Q4_K_M | estimated18 tok/s | 256K (full) | M4 page |
| MacBook Air M2 24 GB refurb | Runs | Q4_K_M | estimated15 tok/s | 256K (full) | — |
| MacBook Air M3 24 GB refurb | Runs | Q4_K_M | estimated15 tok/s | 256K (full) | M3 page |
| MacBook Pro | |||||
| MacBook Pro M6 16 GB | Runs | 4-bit | community30 tok/s | 148K | M6 page |
| MacBook Pro M5 Pro 24 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| MacBook Pro M6 24 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| MacBook Pro M6 32 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| MacBook Pro M5 Pro 48 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| MacBook Pro M5 Max 64 GB | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | M5 Max page |
| MacBook Pro M5 Pro 64 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| MacBook Pro M5 Max 128 GB | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | M5 Max page |
| MacBook Pro M3 Max 64 GB refurb | Runs | Q4_K_M | estimated63 tok/s | 256K (full) | — |
| MacBook Pro M3 Max 128 GB refurb | Runs | Q4_K_M | estimated63 tok/s | 256K (full) | — |
| Mac mini | |||||
| Mac mini M6 16 GB | Runs | 4-bit | community30 tok/s | 148K | M6 page |
| Mac mini M5 Pro 24 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| Mac mini M6 24 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| Mac mini M6 32 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| Mac mini M5 Pro 48 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| Mac mini M5 Pro 64 GB | Runs | Q4_K_M | estimated56 tok/s | 256K (full) | — |
| Mac mini M2 Pro 16 GB refurb | Runs | Q4_K_M | estimated35 tok/s | 113K | — |
| Mac mini M4 16 GB used | Runs | Q4_K_M | estimated18 tok/s | 113K | M4 page |
| Mac mini M4 Pro 24 GB used | Runs | Q4_K_M | community39 tok/s | 256K (full) | M4 Pro page |
| Mac mini M4 24 GB used | Runs | Q4_K_M | estimated18 tok/s | 256K (full) | M4 page |
| Mac mini M2 Pro 32 GB refurb | Runs | Q4_K_M | estimated35 tok/s | 256K (full) | — |
| Mac mini M4 Pro 48 GB used | Runs | Q4_K_M | community39 tok/s | 256K (full) | M4 Pro page |
| Mac mini M4 Pro 64 GB used | Runs | Q4_K_M | community39 tok/s | 256K (full) | M4 Pro page |
| iMac | |||||
| iMac M6 16 GB | Runs | 4-bit | community30 tok/s | 148K | M6 page |
| iMac M6 24 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| iMac M6 32 GB | Runs | Q4_K_M | estimated30 tok/s | 256K (full) | M6 page |
| Mac Studio | |||||
| Mac Studio M5 Max 36 GB | Runs | Q4_K_M | estimated63 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Max 48 GB | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Max 64 GB | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Ultra 96 GB | Runs | 4-bit | community167 tok/s | 256K (full) | M5 Ultra page |
| Mac Studio M5 Max 128 GB | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | M5 Max page |
| Mac Studio M5 Ultra 256 GB | Runs | 4-bit | community167 tok/s | 256K (full) | M5 Ultra page |
| Mac Studio M5 Ultra 512 GB | Runs | 4-bit | community167 tok/s | 256K (full) | M5 Ultra page |
| Mac Studio M2 Max 32 GB refurb | Runs | Q4_K_M | estimated63 tok/s | 256K (full) | — |
| Mac Studio M1 Ultra 64 GB refurb | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | — |
| Mac Studio M4 Max 64 GB used | Runs | Q4_K_M | estimated73 tok/s | 256K (full) | — |
| Mac Studio M2 Max 64 GB refurb | Runs | Q4_K_M | estimated63 tok/s | 256K (full) | — |
| Mac Studio M1 Ultra 128 GB refurb | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | — |
| Mac Studio M2 Ultra 128 GB refurb | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | — |
| Mac Studio M4 Max 128 GB used | Runs | Q4_K_M | estimated73 tok/s | 256K (full) | — |
| Mac Studio M2 Ultra 192 GB refurb | Runs | Q4_K_M | estimated81 tok/s | 256K (full) | — |
| Mac Studio M3 Ultra 256 GB used | Runs | Q4_K_M | community81 tok/s | 256K (full) | M3 Ultra page |
| Mac Studio M3 Ultra 512 GB refurb | Runs | Q4_K_M | community81 tok/s | 256K (full) | M3 Ultra page |
Context memory is computed from its config.json (hybrid: full attention on 8 of 32 layers; Gated DeltaNet layers keep a constant state): about 2.1 GB of 16-bit KV cache at 64K tokens. A q8 cache (llama.cpp -ctk q8_0 -ctv q8_0, Ollama OLLAMA_KV_CACHE_TYPE=q8_0) halves that part. Runtimes also grow a compute buffer with context, so a fit within about 1 GB of the limit is optimistic.
macOS lets the GPU wire only part of unified memory by default (Metal’s recommended working set; M5 and M6
Macs run macOS 26 or later, whose default is higher), and Qwen 3.5 9B needs about 7 GB. On the Macs marked
Raise GPU limit it fits in RAM but not inside that default. The default
is a performance threshold, not a hard wall: past it, the runtime moves part of it to the CPU or shortens the context (llama.cpp, Ollama), runs it slowly (MLX), or stops with an out-of-memory error. One command raises the limit until the next restart
(sudo sysctl iogpu.wired_limit_mb=0 restores the default sooner); it leaves macOS a reserve, and other
apps get less memory while the model is loaded. On the Macs marked Over GPU limit even that raise falls short: the model fits the Mac’s RAM on paper, but a smaller quant or a bigger Mac is the realistic route.
Weights are 9B parameters at each format’s effective size (the index’s GB-per-billion table), plus about 2 GB for the runtime and a short context; longer contexts add the KV cache computed from the model’s config.json at 16 bits. The first row is the catalog’s own figure (about 7 GB at a 4K context), the one every verdict on this page uses. “Smallest Mac” is the smallest memory size among the Macs Apple sells new that holds it at a 4K context inside that Mac’s default GPU limit. Estimates, not measurements.
| Build | Weights | RAM at 4K | at 32K | at 64K | Smallest Mac |
|---|---|---|---|---|---|
| As catalogued | — | 7 GB | 8 GB | 9 GB | 16 GB |
| Q2_K | 3.1 GB | 5 GB | 6 GB | 7 GB | 16 GB |
| Q3_K_M | 4.4 GB | 6 GB | 7 GB | 8 GB | 16 GB |
| Q4_K_M | 5.1 GB | 7 GB | 8 GB | 9 GB | 16 GB |
| Q5_K_M | 6.4 GB | 8 GB | 9 GB | 10 GB | 16 GB |
| Q6_K | 7.4 GB | 9 GB | 10 GB | 11 GB | 16 GB |
| Q8_0 | 9.5 GB | 12 GB | 12 GB | 14 GB | 16 GB |
| F16 | 18.0 GB | 20 GB | 21 GB | 22 GB | 32 GB |
Every published (filled square) and community (hollow circle) run the index holds for Qwen 3.5 9B: 0 published, 6 community, each linked to its source. LLMCheck runs no tests of its own. One row is one measurement session; on each chip the table above shows the lower median of the eligible runs (plain decode at 5 bits or fewer), and a binned or “not an anchor” run is shown but never used as a reference.
| Chip | Runtime | Quant | Decode | Context | Date | Source |
|---|---|---|---|---|---|---|
| M5 Ultra | MLX | 4-bit | community167 tok/s | 512-token random prompt, 128 generated, empty context, mean of 2 | 2026-10-15 | bohemiansilicon.com |
| M3 Ultra | MLX | 4-bit | community114.1 tok/s | 75-token chat prompt, three greedy 256-token continuations | 2026-11-07 | github.com |
| M3 Ultra | llama.cpp | Q4_K_M | community81 tok/s | 75-token chat prompt, three greedy 256-token continuations | 2026-11-07 | github.com |
| M4 Pro | MLX | 4-bit | community50.2 tok/s | 75-token chat prompt, three greedy 256-token continuations | 2026-11-07 | github.com |
| M4 Pro | llama.cpp | Q4_K_M | community39.1 tok/s | 75-token chat prompt, three greedy 256-token continuations | 2026-11-07 | github.com |
| M6 | MLX | 4-bit | community29.8 tok/s | 1k | 2026-11-03 | omlx.ai binned |
The quickest start is Ollama. Pull the model by its tag (checked on ollama.com, 2026-11-14; the file is about 7.6 GB):
Prefer a window to a terminal? LM Studio downloads the same model with one click. · model card →
On the Mac mini M6 (16 GB), the smallest current Mac that runs it inside the default GPU limit, the index figure is 30 tok/s (community run) at 4-bit; keep the context at or under 148K there, or use a q8 KV cache. The qwen3.5:9b file fits there.
Step-by-step: the Ollama install guide.
Mac mini M6 (16 GB) — smallest current Mac that runs it and cheapest at 15+ tok/s, $899 new (Apple list price, checked 2026-11-14), 30 tok/s (community run).
As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence the verdicts.
Yes. Qwen 3.5 9B needs about 7 GB of unified memory at a short (4K-token) context, so it runs on 53 of the 54 Mac configurations in the LLMCheck index. On 1 more it fits the RAM, but not even a safely raised GPU limit. The smallest Mac Apple sells new that runs it is the Mac mini M6 (16 GB).
About 7 GB at a 4K context: the weights plus about 2 GB for the runtime. Context adds KV cache: about 8 GB in total at 32K and 10 GB at 64K with the default 16-bit cache (about 9 GB at 64K with a q8 cache). These are estimates from the model’s config.json, not measurements.
On an M5 Max, the chip the leaderboard quotes it on, the index figure is ~81 tok/s (estimated) at Q4_K_M. The best measured figure is 167 tok/s on an M5 Ultra 256 GB (community run, 4-bit). Every Mac’s figure is in the table on this page with its provenance: published, community run or estimated.
Among Macs Apple sells new, the Mac mini M6 (16 GB), $899 new (Apple list price, checked 2026-11-14), is the cheapest that reaches 15 tok/s with Qwen 3.5 9B, at 30 tok/s (community run).
On MacBook Air M1 8 GB (used) even a safely raised limit is short: expect heavy swapping, a CPU split or an out-of-memory error. From 16 GB up, every Mac Apple sells new holds it inside the default limit.
Q4_K_M (about 5 GB of weights) is the common 4-bit GGUF choice. Q8_0 (about 10 GB) is closer to full quality and needs a 16 GB Mac at the default GPU limit. Q3_K_M and Q2_K shrink it to about 4 and 3 GB at a quality cost. The index’s reference figure (81 tok/s on an M5 Max) is for its Q4_K_M build.
Its own window is 256K tokens. At macOS’s default GPU limit, the Mac mini M6 (16 GB) holds about 148K. The full window needs about 16 GB, which a 24 GB Mac holds at the default limit. A q8 KV cache roughly doubles these lengths.
LLMCheck. (2026). Can I run Qwen 3.5 9B on a Mac? llmcheck.net. https://llmcheck.net/can-i-run/qwen-35-9b/ (updated ). CC BY 4.0.