Can I run Qwen3 8B on the Apple M4 Pro?
Runs greatQ4_K_M fits in unified memory (6.8 of 16.8 GB usable by the GPU): Runs great, est. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%).
24 GB unified memory · 273.0 GB/s · FP16 8.1 TFLOPS. Assumes 8k context and an f16 KV cache. Verdicts are shown for each memory size.
Verdict by memory size
The page uses the 24 GB configuration; here is every size.
- 24 GBRuns greatest. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%)(used on this page)
- 48 GBRuns greatest. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%)
- 64 GBRuns greatest. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%)
Every quant of Qwen3 8B on the Apple M4 Pro
| Quant | File size | Verdict | Speed | Memory | Runs as | Notes |
|---|---|---|---|---|---|---|
| Q2_K | 3.28 GB | Runs great | est. 46.2 tok/s37.0–55.5calibrated estimate ±20% | 5.1 / 16.8 GB | Unified memory | — |
| IQ4_XS | 4.58 GB | Runs great | est. 35.8 tok/s28.7–43.0calibrated estimate ±20% | 6.4 / 16.8 GB | Unified memory | — |
| Q4_K_Mbaseline | 5.03 GB | Runs great | est. 33.3 tok/s26.6–39.9calibrated estimate ±20% | 6.8 / 16.8 GB | Unified memory | — |
| Q8_0 | 8.71 GB | Runs well | est. 20.9 tok/s16.7–25.1calibrated estimate ±20% | 10.5 / 16.8 GB | Unified memory | — |
Models with a thinking mode spend extra tokens thinking, so their speed thresholds are 1.5× stricter (30 / 12 / 3 tok/s). With thinking turned off, a model may do one grade better.
Where the memory goes at Q4_K_M
- Weights
- 5.0 GB
- KV cache
- 1.2 GB
- Compute buffer
- 0.6 GB
- Free
- 10.0 GB
- Not usable by the GPU
- 7.2 GB
Context length vs. KV cache
| Context | KV f16 | KV q8_0 | KV q4_0 |
|---|---|---|---|
| 4k | 6.2 GBRuns great est. 36.8 tok/s29.5–44.2calibrated estimate ±20% | 5.9 GBRuns great est. 38.8 tok/s31.0–46.5calibrated estimate ±20% | 5.7 GBRuns great est. 39.9 tok/s31.9–47.9calibrated estimate ±20% |
| 8k | 6.8 GBRuns great est. 33.3 tok/s26.6–39.9calibrated estimate ±20% | 6.3 GBRuns great est. 36.6 tok/s29.2–43.9calibrated estimate ±20% | 5.9 GBRuns great est. 38.6 tok/s30.9–46.3calibrated estimate ±20% |
| 16k | 8.1 GBRuns well est. 27.9 tok/s22.3–33.4calibrated estimate ±20% | 7.0 GBRuns great est. 32.8 tok/s26.3–39.4calibrated estimate ±20% | 6.4 GBRuns great est. 36.3 tok/s29.0–43.5calibrated estimate ±20% |
| 32k | 10.7 GBRuns well est. 21.0 tok/s16.8–25.2calibrated estimate ±20% | 8.4 GBRuns well est. 27.2 tok/s21.8–32.7calibrated estimate ±20% | 7.2 GBRuns great est. 32.4 tok/s25.9–38.9calibrated estimate ±20% |
S = Runs great · A = Runs well · B = Runs slowly · F = Won't run
At Q4_K_M the verdict stays Runs great up to 8k context with an f16 KV cache, and up to 16k with q8_0.
Estimated speed
- Generation
- est. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%)
- Prompt processing
- about 382 tok/s (229–535, rough estimate ±40%)
Measured on this exact combination
No public measurement for this exact combination yet; the numbers above are estimates.
If this is not enough
A larger model that also runs well on the Apple M4 Pro
Runs wellMistral Small 3.2 24B: Runs well, est. 13.2 tok/s (10.6–15.9, calibrated estimate ±20%)
How to run it
Commands for the Q4_K_M file. Both tools download from Hugging Face on first run.
File: Qwen3-8B-Q4_K_M.gguf (5.03 GB) from unsloth/Qwen3-8B-GGUF.
llama-server -hf unsloth/Qwen3-8B-GGUF:Q4_K_M -c 8192 -ngl all -fa on- -ngl all loads every layer on the GPU.
- -fa on enables flash attention (needed for KV cache quantization).
- --cache-type-k q8_0 --cache-type-v q8_0 shrinks the KV cache to 0.65 GB at 8k context.
Terminal:
OLLAMA_CONTEXT_LENGTH=8192 ollama serve
ollama run hf.co/unsloth/Qwen3-8B-GGUF:Q4_K_M- Ollama defaults to 4k context on GPUs under 24 GB. Set OLLAMA_CONTEXT_LENGTH when starting the server, or /set parameter num_ctx inside the chat.
/set parameter num_ctx 8192
Related pages
Guide: Running local LLMs on a Mac (M4, M5): how much unified memory?
Other models on the Apple M4 Pro
Qwen3 8B on other hardware
Frequently asked questions
Can I run Qwen3 8B on the Apple M4 Pro?
Q4_K_M fits in unified memory (6.8 of 16.8 GB usable by the GPU): Runs great, est. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%). Q4_K_M needs 6.8 GB at 8k context; the GPU can use 16.8 GB of the unified memory.
How much context can Qwen3 8B use on the Apple M4 Pro?
At Q4_K_M the verdict stays Runs great up to 8k context with an f16 KV cache (1.2 GB of KV) and up to 16k with a q8_0 KV cache (1.3 GB). The model supports up to 32k.
Which quant should I use, and how fast is it?
Q4_K_M (5.03 GB) is the recommended balance: Runs great, est. 33.3 tok/s (26.6–39.9, calibrated estimate ±20%). It is also the largest tracked file that gets this verdict.
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.