The verdict is the Mac mini M6 (24 GB): it is cheaper and at least as fast. The LLMCheck index puts Recursive Flint at ~71 tok/s (estimated) on it, against ~67 tok/s (estimated) on the MacBook Air M5 (24 GB), and both fit 63 of the 130 models in the index. It saves $400 (Apple list, 14 Nov 2026).
| MacBook Air M5 (24 GB) | Mac mini M6 (24 GB) | |
|---|---|---|
| Chip | M5 | M6 |
| GPU cores | 10-core | 12-core |
| Memory bandwidth | 153 GB/s | 170 GB/s |
| Unified memory | 24 GB | 24 GB |
| Price | $1,499 | $1,099 |
| Price per GB | $62 | $46 |
| Models that fit (of 130) | 63 | 63 |
| Best local LLM here | Qwen 3.6-35B-A3B estimated59 tok/s | Recursive Flint estimated71 tok/s |
| Recursive Flint | estimated67 tok/s | estimated71 tok/s |
| Price per tok/s on Recursive Flint | $22 | $15 |
| Form factor | a fanless laptop with a battery and a built-in display | a small desktop (bring your own display) |
| Where to buy | New · sold new by Apple (Amazon stock not verified) | New · Amazon stocks this configuration |
Every speed below is the LLMCheck index’s figure for that exact configuration — the same number its MacBook Air M5 (24 GB) and Mac mini M6 (24 GB) pages print, with its provenance mark. Across the 60 shared models with a figure on both, the Mac mini M6 column is a median ×1.10 of the MB Air M5 column.
| Model | Size | MB Air M5 tok/s | Mac mini M6 tok/s | Ratio |
|---|---|---|---|---|
| Qwen 3.6-35B-A3B | 35B MoE | estimated59 | community59 | ×1.00 |
| Qwen3.8-27B | 27.8B | estimated8 slow | community10 slow | ×1.25 |
| Nanbeige4.3-33B | 33.1B | estimated6 slow | estimated7 slow | ×1.17 |
| Recursive Flint | 39.8B MoE | estimated67 | estimated71 | ×1.06 |
| Qwen 3.6-27B | 27B | estimated8 slow | estimated9 slow | ×1.12 |
| KAT-Coder-V2.5 | 35B MoE | estimated44 | estimated47 | ×1.07 |
| LFM2.5-24B-A2B | 24B MoE | community74 | estimated79 | ×1.07 |
| Muse Glimmer 30B | 30B | estimated6 slow | estimated6 slow | ×1.00 |
| MiniCPM-V 4.7 35B-A3B | 35.2B MoE | estimated52 | estimated54 | ×1.04 |
| Nemotron 3.5 Lightning | 30B MoE | estimated41 | community45 | ×1.10 |
| Gemma 4 31B | 31B | estimated7 slow | community8 slow | ×1.14 |
| Gemma 4 26B-A4B | 26B MoE | estimated48 | community49 | ×1.02 |
| Raon-1 mini preview | 28.8B MoE | estimated54 | estimated58 | ×1.07 |
| GLM-4.7-Flash | 31B MoE | estimated45 | community47 | ×1.04 |
| North Mini Code 1.0 | 30.5B MoE | community34 | estimated37 | ×1.09 |
Nothing: both fit the same 63 of the 130 models in the LLMCheck index, because both have 24 GB. The difference is speed and price.
At Apple’s list prices the MacBook Air M5 (24 GB) costs $62 per GB of unified memory and the Mac mini M6 (24 GB) $46. On Recursive Flint — the reference model here, which both run — that is $22 and $15 per tok/s of decode speed: the Mac mini M6 (24 GB) buys more speed per dollar.
Bottom line: The Mac mini M6 (24 GB): cheaper and at least as fast.
The MacBook Air M5 (24 GB) is sold new by Apple; Amazon stock of the 24 GB configuration is not verified, so check the listing’s memory.
As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence the verdicts.
Each speed is decode (generation) tok/s from the LLMCheck index for that exact configuration: the lower-median measured run on that chip where one exists (a filled square is a published figure, a hollow circle a community run), otherwise an estimate from the memory-bandwidth model (hollow diamond, printed with a tilde). LLMCheck runs no tests of its own. Prompt processing (prefill) depends on GPU compute rather than bandwidth, and the index does not estimate it.
The verdict is the Mac mini M6 (24 GB): it is cheaper and at least as fast. The LLMCheck index puts Recursive Flint at ~71 tok/s (estimated) on it, against ~67 tok/s (estimated) on the MacBook Air M5 (24 GB), and both fit 63 of the 130 models in the index. It saves $400 (Apple list, 14 Nov 2026).
Decode speed on Apple Silicon follows memory bandwidth: the Mac mini M6 (24 GB) reads memory at 170 GB/s, the MacBook Air M5 (24 GB) at 153 GB/s. The LLMCheck index puts Recursive Flint at ~71 tok/s (estimated) against ~67 tok/s (estimated). Across the 60 shared models with a figure on both, the median gain is 1.10x.
Yes. Both fit the same 63 of the 130 models in the LLMCheck index.
It costs $1,499 against $1,099 (Apple list prices, checked 14 Nov 2026; the MacBook Air M5 (24 GB) price includes 512GB of storage, the Mac mini M6 (24 GB) price 256GB). For that it gives ~67 tok/s (estimated) against ~71 tok/s (estimated) on Recursive Flint. For local LLMs the cheaper Mac is the better buy; pay more for its form factor (a fanless laptop with a battery and a built-in display), not for speed.
On the MacBook Air M5 (24 GB): Qwen 3.6-35B-A3B at ~59 tok/s (estimated). On the Mac mini M6 (24 GB): Recursive Flint at ~71 tok/s (estimated). These are the headline picks of the LLMCheck Best by Mac pages: the best-ranked model that reaches 15 tok/s on that Mac and has a standard Mac runtime.