Can I run Qwen3 14B on the Apple M4 Pro?
Runs wellQ4_K_M fits in unified memory (10.9 of 16.8 GB usable by the GPU): Runs well, est. 20.1 tok/s (16.0–24.1, 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 wellest. 20.1 tok/s (16.0–24.1, calibrated estimate ±20%)(used on this page)
- 48 GBRuns wellest. 20.1 tok/s (16.0–24.1, calibrated estimate ±20%)
- 64 GBRuns wellest. 20.1 tok/s (16.0–24.1, calibrated estimate ±20%)
Every quant of Qwen3 14B on the Apple M4 Pro
| Quant | File size | Verdict | Speed | Memory | Runs as | Notes |
|---|---|---|---|---|---|---|
| Q2_K | 5.75 GB | Runs well | est. 29.3 tok/s23.4–35.1calibrated estimate ±20% | 7.7 / 16.8 GB | Unified memory | — |
| IQ4_XS | 8.14 GB | Runs well | est. 21.9 tok/s17.5–26.3calibrated estimate ±20% | 10.1 / 16.8 GB | Unified memory | — |
| Q4_K_Mbaseline | 9.00 GB | Runs well | est. 20.1 tok/s16.0–24.1calibrated estimate ±20% | 10.9 / 16.8 GB | Unified memory | — |
| Q8_0 | 15.70 GB | Runs slowly | est. 8.0 tok/s5.6–10.4theoretical estimate ±30% | 17.0 GB RAM | CPU only | — |
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
- 9.0 GB
- KV cache
- 1.3 GB
- Compute buffer
- 0.6 GB
- Free
- 5.9 GB
- Not usable by the GPU
- 7.2 GB
Context length vs. KV cache
| Context | KV f16 | KV q8_0 | KV q4_0 |
|---|---|---|---|
| 4k | 10.2 GBRuns well est. 21.5 tok/s17.2–25.7calibrated estimate ±20% | 9.9 GBRuns well est. 22.2 tok/s17.7–26.6calibrated estimate ±20% | 9.7 GBRuns well est. 22.6 tok/s18.1–27.1calibrated estimate ±20% |
| 8k | 10.9 GBRuns well est. 20.1 tok/s16.0–24.1calibrated estimate ±20% | 10.3 GBRuns well est. 21.3 tok/s17.1–25.6calibrated estimate ±20% | 10.0 GBRuns well est. 22.1 tok/s17.7–26.5calibrated estimate ±20% |
| 16k | 12.3 GBRuns well est. 17.8 tok/s14.2–21.3calibrated estimate ±20% | 11.1 GBRuns well est. 19.9 tok/s15.9–23.9calibrated estimate ±20% | 10.4 GBRuns well est. 21.2 tok/s17.0–25.5calibrated estimate ±20% |
| 32k | 15.2 GBRuns well est. 14.4 tok/s11.6–17.3calibrated estimate ±20% | 12.7 GBRuns well est. 17.5 tok/s14.0–21.0calibrated estimate ±20% | 11.3 GBRuns well est. 19.7 tok/s15.8–23.6calibrated estimate ±20% |
S = Runs great · A = Runs well · B = Runs slowly · F = Won't run
At Q4_K_M the verdict stays Runs well up to 32k context with an f16 KV cache, and up to 32k with q8_0.
Estimated speed
- Generation
- est. 20.1 tok/s (16.0–24.1, calibrated estimate ±20%)
- Prompt processing
- about 216 tok/s (129–302, 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%)
Qwen3 14B on the Apple M4 Max (32-core GPU)
Runs greatQ4_K_M: Runs great, est. 30.1 tok/s (24.1–36.2, calibrated estimate ±20%)
How to run it
Commands for the Q4_K_M file. Both tools download from Hugging Face on first run.
File: Qwen3-14B-Q4_K_M.gguf (9.00 GB) from unsloth/Qwen3-14B-GGUF.
llama-server -hf unsloth/Qwen3-14B-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 0.72 GB at 8k context.
Terminal:
OLLAMA_CONTEXT_LENGTH=8192 ollama serve
ollama run hf.co/unsloth/Qwen3-14B-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 14B on other hardware
Frequently asked questions
Can I run Qwen3 14B on the Apple M4 Pro?
Q4_K_M fits in unified memory (10.9 of 16.8 GB usable by the GPU): Runs well, est. 20.1 tok/s (16.0–24.1, calibrated estimate ±20%). Q4_K_M needs 10.9 GB at 8k context; the GPU can use 16.8 GB of the unified memory.
How much context can Qwen3 14B use on the Apple M4 Pro?
At Q4_K_M the verdict stays Runs well up to 32k context with an f16 KV cache (5.4 GB of KV) and up to 32k with a q8_0 KV cache (2.9 GB). The model supports up to 32k.
Which quant should I use, and how fast is it?
Q4_K_M (9.00 GB) is the recommended balance: Runs well, est. 20.1 tok/s (16.0–24.1, 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.