Apple · M2 · Laptop
Best AI Models for MacBook Air 13" M2 (16 GB)
Runs 9B models at 11 tok/s; 14B only just fits in 12.5 GB, and 27–31B doesn't fit.
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Qwen3.5 9BAlibabaHow to run → Q4_K_M · 6.6 GB | 11.0 tok/s | |
| 2 | Gemma 3 4B ITGoogle Q4_K_M · 5.0 GB | 24.6 tok/s | |
| 3 | Q4_K_M · 2.3 GB | 85.4 tok/s | |
| 4 | Gemma 3 1B ITGoogle Q4_K_M · 2.1 GB | 106 tok/s | |
| 5 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB | 21.5 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 5.0 GB | 24.6 tok/s | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB | 11.0 tok/s | |
| 3 | Q4_K_M · 2.3 GB | 85.4 tok/s | overall |
| 4 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB | 21.5 tok/s | |
| 5 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB | 55.1 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 2.3 GB | 85.4 tok/s | |
| 2 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB | 21.5 tok/s | |
| 3 | Ling 3.0 TinyInclusion AI Q4_K_M · 5.5 GB | 55.1 tok/s | |
| 4 | Gemma 4 E2B ITGoogle Q4_K_M · 3.5 GB | 20.4 tok/s | |
| 5 | LFM2.5 8B A1BLiquid AI Q4_K_M · 5.6 GB | 49.1 tok/s |
| # | Model | Speed | Score |
|---|---|---|---|
| 1 | Q4_K_M · 5.0 GB | 24.6 tok/s | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB | 11.0 tok/s | |
| 3 | Qwen3.5 4BAlibaba Q4_K_M · 3.4 GB | 21.5 tok/s | |
| 4 | Cosmos Reason2 2BNVIDIA Q4_K_M · 3.5 GB | 43.5 tok/s | |
| 5 | LocateAnything 3BNVIDIA Q4_K_M · 3.2 GB | 26.2 tok/s |
The MacBook Air 13" M2 is a 16 GB laptop with more memory than speed: it runs 9B models at 11 tok/s. macOS keeps about 3.5 GB, leaving 12.5 GB for models. Prompt reading is its weak spot: a 4K-token prompt waits 55 s before the first word on Qwen3.5 9B. M2 generation, released 2022; the newer M3 writes 2% faster with 1.0× the compute.
MacBook Air 13" M2 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 7.4 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 →
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 Qwen3.6 35B A3B.
Output speed depends on memory bandwidth (100 GB/s). Grey rows need more than the 12.5 GB available.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB11.0tok/s
- 7.4tok/s
- 27–31BQwen3.8 27B · does not fitneeds 17.9 GB
- 35B MoEQwen3.6 35B A3B · does not fitneeds 22.0 GB
- 70BLlama 3.3 70B Instruct · does not fitneeds 51.3 GB
- 120B MoEGPT OSS 120B · does not fitneeds 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: 24.9 to 44.9 tok/s since the M1. Compute, which sets how fast prompts are read, grew 2.5×. Indexed to the M1; speeds on Qwen3.6 35B A3B, same 16 GB.
All 918 models on the MacBook Air 13" M2
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 |
|---|---|---|---|---|---|---|
| Q4_K_M | 6.4 GB | 12.3 t/s | 33K | B51, Decent | ||
DeepHat V1 7BChatCode | Q4_K_M | 5.8 GB | 14.0 t/s | 33K | B53, Decent | |
C4ai Command R7b 12 2024ChatFunctions | Q4_K_M | 7.9 GB | 13.2 t/s | — | B52, Decent | |
| Q4_K_M | 1.7 GB | 80.5 t/s | 33K | S88, Runs great | ||
Qwen3 VL 2B ThinkingVisionFunctions | Q4_K_M | 3.5 GB | 38.7 t/s | 262K | A75, Runs well | |
Qwen2.5 VL 7B InstructVision | Q4_K_M | 6.2 GB | 13.0 t/s | 128K | B52, Decent | |
| Q3_K_M | 11.1 GB | 25.4 t/s | 131K | B53, Decent | ||
Pythia 410MChat | Q4_K_M | 0.3 GB | 212.1 t/s | 2K | S100, Runs great | |
| Q3_K_M | 10.8 GB | 28.5 t/s | 131K | B62, Decent | ||
Gemma 3 4B ITVision | Q4_K_M | 5 GB | 24.6 t/s | — | A65, Runs well | |
Gpt2 MediumChat | Q4_K_M | 0.3 GB | 280.0 t/s | 1K | S100, Runs great | |
TinyStories 1MChat | BF16 | 0.0 GB | 7000.0 t/s | 2K | S100, Runs great | |
Sarvam 1Chat | Q4_K_M | 2.8 GB | 34.1 t/s | 8K | A73, Runs well | |
Param 1 5BChat | BF16 | 10.6 GB | 6.7 t/s | 4K | C38, Usable | |
| BF16 | 0.6 GB | 1166.7 t/s | 450K | S100, Runs great |
Specs
- Chip
- M2
- CPU cores
- 8
- GPU cores
- 10
- Unified memory
- 16 GB
- For models
- 12.5 GB (macOS keeps 3.5 GB)
- Memory bandwidth
- 100 GB/s
- FP16 (dense)
- 7.2 TFLOPS
- Neural Engine
- 16 TOPS
- Released
- 2022-07-15
MacBook Air 13" M2 FAQ
How much memory is available for AI on the MacBook Air 13" M2 (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 13" M2 (16 GB) can run?
GPT OSS 20B (20.9B) at Q4_K_S uses 12.4 GB, 99% of the available memory, and runs at about 25.3 tok/s.
Can the MacBook Air 13" M2 (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 13" M2 (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 13" M2 (16 GB) take to read a long prompt?
On Qwen3.5 9B, about 55 s for a 4K-token prompt and 7.4 min for a 32K-token document before the first word appears.
Is renting a GPU cheaper than buying the MacBook Air 13" M2?
A Tesla V100 (16 GB) rents from $0.10/hr.