Apple · M4 · Laptop
Best AI Models for MacBook Air 15" M4 (16 GB)
Runs 9B models at 13 tok/s; 14B only just fits in 12.5 GB, and 27–31B doesn't fit.
| # | Model | Speed | llmrun Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB · 13.2 t/s | 13.2 tok/s | Q4_K_M · 6.6 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 25.8 t/s | 25.8 tok/s | Q4_K_M · 3.4 GB | |
| 3 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 29.6 t/s | 29.6 tok/s | Q4_K_M · 5.0 GB | |
| 4 | Q4_K_M · 7.9 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 7.9 GB | |
| 5 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 62.8 t/s | 62.8 tok/s | Q4_K_M · 5.5 GB |
| # | Model | Speed | Coding Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Q4_K_M · 7.9 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 7.9 GB | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB · 25.8 t/s | 25.8 tok/s | overall | Q4_K_M · 3.4 GB |
| 3 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB · 15.2 t/s | 15.2 tok/s | overall | Q4_K_M · 7.6 GB |
| 4 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 62.8 t/s | 62.8 tok/s | Q4_K_M · 5.5 GB | |
| 5 | Gemma 4 E4B ITGoogle Q4_K_M · 5.6 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 5.6 GB |
| # | Model | Speed | Reasoning Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Qwen3 8BAlibabaHow to run → Q4_K_M · 7.6 GB · 15.2 t/s | 15.2 tok/s | Q4_K_M · 7.6 GB | |
| 2 | Q4_K_M · 7.9 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 7.9 GB | |
| 3 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 29.6 t/s | 29.6 tok/s | Q4_K_M · 5.0 GB | |
| 4 | Qwen2.5 7B InstructAlibaba Q4_K_M · 5.8 GB · 16.8 t/s | 16.8 tok/s | Q4_K_M · 5.8 GB | |
| 5 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 62.8 t/s | 62.8 tok/s | Q4_K_M · 5.5 GB |
| # | Model | Speed | Long context Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 25.8 t/s | 25.8 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Q4_K_M · 7.9 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 7.9 GB | |
| 3 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 62.8 t/s | 62.8 tok/s | Q4_K_M · 5.5 GB | |
| 4 | Gemma 4 E4B ITGoogle Q4_K_M · 5.6 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 5.6 GB | |
| 5 | Gemma 4 E2B ITGoogle Q4_K_M · 3.5 GB · 24.5 t/s | 24.5 tok/s | Q4_K_M · 3.5 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 25.8 t/s | 25.8 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB · 62.8 t/s | 62.8 tok/s | Q4_K_M · 5.5 GB | |
| 3 | Gemma 4 E2B ITGoogle Q4_K_M · 3.5 GB · 24.5 t/s | 24.5 tok/s | Q4_K_M · 3.5 GB | |
| 4 | Gemma 4 E4B ITGoogle Q4_K_M · 5.6 GB · 15.8 t/s | 15.8 tok/s | Q4_K_M · 5.6 GB | |
| 5 | Qwen3 8BAlibaba Q4_K_M · 7.6 GB · 15.2 t/s | 15.2 tok/s | Q4_K_M · 7.6 GB |
| # | Model | Speed | Vision Score | Quant · Memory |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 25.8 t/s | 25.8 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB · 29.6 t/s | 29.6 tok/s | Q4_K_M · 5.0 GB | |
| 3 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 13.2 t/s | 13.2 tok/s | Q4_K_M · 6.6 GB | |
| 4 | MiniCPM V 4.7 1BOpenBMB Q4_K_M · 1.6 GB · 65.6 t/s | 65.6 tok/s | Q4_K_M · 1.6 GB | |
| 5 | LFM3 Omni 2BLiquid AI Q4_K_M · 2.4 GB · 49.4 t/s | 49.4 tok/s | Q4_K_M · 2.4 GB |
The MacBook Air 15" M4 is a mid-range 16 GB laptop: it runs 9B models at 13 tok/s. Among 16 GB machines, only the M1 Pro, M2 Pro and M5 write faster. macOS keeps about 3.5 GB, leaving 12.5 GB for models. Prompt reading is its weak spot: a 4K-token prompt waits 57 s before the first word on Qwen3.5 9B. M4 generation, released 2025; the newer M5 writes 28% faster with 2.3× the compute.
MacBook Air 15" M4 strengths and weaknesses
Weaknesses
- 27–31B models don't fit: Qwen3.8 27B needs 17.9 GB at Q4 vs 12.5 GB usable here
- Little room for context: Qwen3 14B uses 11.9 of 12.5 GB usable, 0.6 GB left
- Slow prompt reading: A 32K-token prompt on Qwen3.5 9B waits 7.6 min for the first token
- No CUDA: vLLM, AWQ, GPTQ and FP8 builds don't run; use GGUF or MLX
Memory for models
12.5 GB of 16 GB27–31B models need 17.9 GB at Q4, more than the 12.5 GB available. The 24 GB config fits Qwen3.8 27B →Or rent an RTX 3090 (24 GB) from $0.20/hr →
Other memory sizes of the same machine first, then machines and GPUs people weigh against it, each with the one reason it is worth a look. Speeds on Qwen3.5 9B.
Output speed depends on memory bandwidth (120 GB/s). Grey rows need more than the 12.5 GB available.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB13.2tok/s
- 8.8tok/s
- 27–31BQwen3.8 27B · not at 4-bitneeds 17.9 GB
- 35B MoEQwen3.6 35B A3B · not at 4-bitneeds 22.0 GB
- 70BLlama 3.3 70B Instruct · not at 4-bitneeds 51.3 GB
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
Chip generations: writing vs compute
M1 → M5Writing speed follows memory bandwidth and barely moved: 7.5 to 16.9 tok/s since the M1. Compute, which sets how fast prompts are read, grew 3.1×. Indexed to the M1; speeds on Qwen3.5 9B, same 16 GB.
All 971 models on the MacBook Air 15" M4
Featured onlyBest quant that fits in 12.5 GB, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
JevK5ChatFunctions | Q4_K_M | 3.2 GB | 28.1 t/s | 262K | A69, Runs well | |
Qwen3 0.6BChatFunctions | Q4_K_M | 1.7 GB | 96.6 t/s | 41K | S90, Runs great | |
InternVL3 1BVision | Q4_K_M | 2.0 GB | 135.5 t/s | — | S96, Runs great | |
Opt 125MChat | BF16 | 0.3 GB | 300.0 t/s | 2K | S100, Runs great | |
Moore LmChat | BF16 | 0.7 GB | 933.3 t/s | — | S100, Runs great | |
Agents A1 4BChatVisionFunctions | Q4_K_M | 3.4 GB | 26.3 t/s | 262K | A67, Runs well | |
MiniCPM5 1BChatFunctions | Q4_K_M | 1.3 GB | 84.8 t/s | 131K | S88, Runs great | |
Olmo 3 7B InstructChatFunctions | Q4_K_M | 8.4 GB | 14.6 t/s | 66K | B54, Decent | |
GPT S 5MChat | BF16 | 0.3 GB | 262.5 t/s | 1K | S100, Runs great | |
| Q4_K_M | 0.0 GB | 8400.0 t/s | 1K | S100, Runs great | ||
| Q4_K_M | 0.0 GB | 8400.0 t/s | 1K | S100, Runs great | ||
Supra 50M ReasoningChatReasoning | Q4_K_M | 0.3 GB | 233.3 t/s | 1K | S100, Runs great | |
CAJAL 4BChat | Q4_K_M | 4.8 GB | 31.8 t/s | — | A71, Runs well | |
MiniCPM5 1B SFTChat | Q4_K_M | 1.3 GB | 84.8 t/s | 131K | S88, Runs great | |
| Q4_K_M | 7.9 GB | 15.8 t/s | — | B55, Decent |
Specs
- Chip
- M4
- CPU cores
- 10
- GPU cores
- 8
- Unified memory
- 16 GB
- For models
- 12.5 GB (macOS keeps 3.5 GB)
- Memory bandwidth
- 120 GB/s
- FP16 (dense)
- 6.9 TFLOPS
- Neural Engine
- 38 TOPS
- Released
- 2025-03-12
MacBook Air 15" M4 FAQ
How much memory is available for AI on the MacBook Air 15" M4 (16 GB)?
About 12.5 GB of the 16 GB. macOS keeps roughly 3.5 GB for itself and other apps.
What is the largest model the MacBook Air 15" M4 (16 GB) can run?
GPT OSS 20B (20.9B) is the largest at 4-bit or better: at Q4_K_S it uses 12.4 GB, 99% of the available memory, and runs at about 29.6 tok/s.
Can the MacBook Air 15" M4 (16 GB) run Qwen3.5 9B?
Yes. Qwen3.5 9B fits at Q4_K_M, using 6.6 GB of the 12.5 GB available, and it is the best model that runs well on this machine.
Can the MacBook Air 15" M4 (16 GB) run 27–31B models?
No. Qwen3.8 27B needs 17.9 GB at Q4, more than the 12.5 GB available.
How long does the MacBook Air 15" M4 (16 GB) take to read a long prompt?
On Qwen3.5 9B, about 57 s for a 4K-token prompt and 7.6 min for a 32K-token document before the first word appears.
Is renting a GPU cheaper than buying the MacBook Air 15" M4?
An RTX 5080 (16 GB) rents from $0.22/hr.