The verdict is the MacBook Pro M5 Max (64 GB), if the $2,000 premium fits: the LLMCheck index’s figures put it a median 1.9× faster across the 69 shared models with a figure on both; on Qwen 3.6-35B-A3B, 130 against 79 tok/s (both community runs). It also fits 6 more of the index’s models.
| M5 Pro (48 GB) | M5 Max (64 GB) | |
|---|---|---|
| Chip | M5 Pro | M5 Max |
| GPU cores | 16- or 20-core | 40-core |
| Memory bandwidth | 307 GB/s | 614 GB/s |
| Unified memory | 48 GB | 64 GB |
| Price | $3,099 | $5,099 |
| Price per GB | $65 | $80 |
| Models that fit (of 130) | 71 | 77 |
| Best local LLM here | LFM2.5-24B-A2B community114 tok/s | Qwen 3.6-35B-A3B community130 tok/s |
| Qwen 3.6-35B-A3B | community79 tok/s | community130 tok/s |
| Price per tok/s on Qwen 3.6-35B-A3B | $39 | $39 |
| Where to buy | New · build-to-order from Apple | New · build-to-order from Apple |
Every speed below is the LLMCheck index’s figure for that exact configuration — the same number its MacBook Pro M5 Pro (48 GB) and MacBook Pro M5 Max (64 GB) pages print, with its provenance mark. Across the 69 shared models with a figure on both, the M5 Max column is a median ×1.86 of the M5 Pro column.
| Model | Size | M5 Pro tok/s | M5 Max tok/s | Ratio |
|---|---|---|---|---|
| LFM2.5-24B-A2B | 24B MoE | community114 | sourced184 | ×1.61 |
| Qwen3.8-27B | 27.8B | sourced17 | community31 | ×1.82 |
| Recursive Flint | 39.8B MoE | community92 | community123 | ×1.34 |
| KAT-Coder-V2.5 | 35B MoE | estimated72 | community89 | ×1.24 |
| Qwen 3.6-35B-A3B | 35B MoE | community79 | community130 | ×1.65 |
| Nanbeige4.3-33B | 33.1B | estimated12 slow | community25 | ×2.08 |
| Qwen 3.6-27B | 27B | estimated15 | estimated30 | ×2.00 |
| Nemotron 3.5 Lightning | 30B MoE | community72 | community117 | ×1.62 |
| North Mini Code 1.0 | 30.5B MoE | community70 | community120 | ×1.71 |
| Gemma 4 26B-A4B | 26B MoE | estimated78 | sourced107 | ×1.37 |
| Raon-1 mini preview | 28.8B MoE | community86 | estimated123 | ×1.43 |
| GLM-4.7-Flash | 31B MoE | estimated73 | estimated107 | ×1.47 |
| Nemotron 3.5 Bolt 74B-A9B caveat | 74B MoE | estimated31 | community58 | ×1.87 |
| Muse Glimmer 30B | 30B | estimated11 slow | sourced27 | ×2.45 |
| Laguna XS 2.1 | 33B MoE | estimated83 | estimated117 | ×1.41 |
A model fits when its minimum RAM (its smallest build of at least 3 bits with a working Mac runtime, plus about 2 GB for the runtime and a short context; 2-bit only where a published Mac run used it) is at most the Mac’s unified memory. The extra memory unlocks these models from the LLMCheck index:
| Model | Size | Min RAM | M5 Max tok/s |
|---|---|---|---|
| Ling-3.0-flash-VL | 124B MoE | 56 GB | estimated53 |
| Kolibri-1 | 78.1B MoE | 50 GB | estimated75 |
| GPT-oss 120B | 117B MoE | 62 GB | estimated92 |
| Mistral Small 4 | 119B MoE | 64 GB | estimated46 |
| GLM-4.5-Air | 106B MoE | 56 GB | estimated27 |
| Llama 4 Scout | 109B MoE | 58 GB | estimated19 |
At Apple’s list prices the M5 Pro (48 GB) costs $65 per GB of unified memory and the M5 Max (64 GB) $80. On Qwen 3.6-35B-A3B — the reference model here, which both run — that is $39 and $39 per tok/s of decode speed: a tie, so the premium buys speed at the same rate.
The $2,000 premium (65% more) buys +65% decode speed on Qwen 3.6-35B-A3B and 6 more models that fit.
Bottom line: The M5 Max (64 GB): a median 1.9× the speed for $2,000 more.
Amazon does not stock either configuration: Apple builds both to order.
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 MacBook Pro M5 Max (64 GB), if the $2,000 premium fits: the LLMCheck index’s figures put it a median 1.9× faster across the 69 shared models with a figure on both; on Qwen 3.6-35B-A3B, 130 against 79 tok/s (both community runs). It also fits 6 more of the index’s models.
Decode speed on Apple Silicon follows memory bandwidth: the MacBook Pro M5 Max (64 GB) reads memory at 614 GB/s, the MacBook Pro M5 Pro (48 GB) at 307 GB/s. The LLMCheck index puts Qwen 3.6-35B-A3B at 130 tok/s (community run) against 79 tok/s (community run). Across the 69 shared models with a figure on both, the median gain is 1.86x.
6 models in the LLMCheck index need more than 48 GB but fit in 64 GB, led by Ling-3.0-flash-VL, Kolibri-1, GPT-oss 120B. Both run the other 71.
It costs $5,099 against $3,099 (Apple list prices, checked 14 Nov 2026; the M5 Max (64 GB) price includes 2TB of storage, the M5 Pro (48 GB) price 1TB). For that it gives 130 tok/s (community run) against 79 tok/s (community run) on Qwen 3.6-35B-A3B and fits 6 more models. Worth it if you use the extra speed or models daily; the cheaper Mac still runs all but 6 of the same models.
On the MacBook Pro M5 Pro (48 GB): LFM2.5-24B-A2B at 114 tok/s (community run). On the MacBook Pro M5 Max (64 GB): Qwen 3.6-35B-A3B at 130 tok/s (community run). 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.