The verdict is the Mac Studio M5 Ultra (256 GB), if the $4,400 premium fits: the LLMCheck index’s figures put it a median 1.6× faster across the 89 shared models with a figure on both; on Qwen 3.6-35B-A3B, 153 against 130 tok/s (both community runs). It also fits 15 more of the index’s models.
| M5 Max (128 GB) | M5 Ultra (256 GB) | |
|---|---|---|
| Chip | M5 Max | M5 Ultra |
| GPU cores | 40-core | 64- or 80-core |
| Memory bandwidth | 614 GB/s | 1,228 GB/s |
| Unified memory | 128 GB | 256 GB |
| Price | $5,099 | $9,499 |
| Price per GB | $40 | $37 |
| Models that fit (of 130) | 93 | 108 |
| Best local LLM here | Qwen 3.6-35B-A3B community130 tok/s | MiMo-V2.6-Flash community77 tok/s |
| Qwen 3.6-35B-A3B | community130 tok/s | community153 tok/s |
| Price per tok/s on Qwen 3.6-35B-A3B | $39 | $62 |
| 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 Mac Studio M5 Max (128 GB) and Mac Studio M5 Ultra (256 GB) pages print, with its provenance mark. Across the 89 shared models with a figure on both, the M5 Ultra column is a median ×1.64 of the M5 Max column.
| Model | Size | M5 Max tok/s | M5 Ultra tok/s | Ratio |
|---|---|---|---|---|
| Qwen 3.6-35B-A3B | 35B MoE | community130 | community153 | ×1.18 |
| Qwen3.8-Flash-Next | 125B MoE | community47 | community57 | ×1.21 |
| Recursive Flint | 39.8B MoE | community123 | estimated131 | ×1.07 |
| DeepSeek V4 Flash | 284B MoE | community39 | estimated45 | ×1.15 |
| DeepSeek V4 Flash Vision-Exp | 305B MoE | estimated39 | estimated46 | ×1.18 |
| GLM-5.3-Flash | 320B MoE | estimated27 | estimated35 | ×1.30 |
| Inkling-Small caveat | 276B MoE | estimated42 | estimated49 | ×1.17 |
| North Mini Code 1.0 | 30.5B MoE | community120 | estimated150 | ×1.25 |
| Nemotron 3.5 Lightning | 30B MoE | community117 | estimated165 | ×1.41 |
| LFM2.5-24B-A2B | 24B MoE | sourced184 | estimated184 | ×1.00 |
| Qwen3.8-27B | 27.8B | community31 | sourced48 | ×1.55 |
| KAT-Coder-V2.5 | 35B MoE | community89 | estimated120 | ×1.35 |
| Qwen3-235B-A22B | 235B MoE | estimated26 | sourced46 | ×1.77 |
| Nanbeige4.3-33B | 33.1B | community25 | estimated52 | ×2.08 |
| Raon-1 mini preview | 28.8B MoE | estimated123 | estimated152 | ×1.24 |
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 Ultra tok/s |
|---|---|---|---|
| MiMo-V2.6-Flash | 309B MoE | 170 GB | community77 |
| MiniMax-M3 | 428B MoE | 162 GB | estimated42 |
| GLM 5.2 | 753B MoE | 256 GB | estimated22 |
| Inkling 2 Small caveat | 594B MoE | 228 GB | estimated36 |
| MiniMax-M3.1-Flash | 428B MoE | 189 GB | community35 |
| GLM-5.3 | 753B MoE | 256 GB | estimated17 |
| Beam 1.0 caveat | 501.3B MoE | 199 GB | estimated27 |
| MiniMax M2.5 | 230B MoE | 130 GB | estimated57 |
| Qwen 3.5 397B-A17B | 397B MoE | 210 GB | community62 |
| Hunyuan Hy3 | 295B MoE | 160 GB | estimated30 |
At Apple’s list prices the M5 Max (128 GB) costs $40 per GB of unified memory and the M5 Ultra (256 GB) $37. On Qwen 3.6-35B-A3B — the reference model here, which both run — that is $39 and $62 per tok/s of decode speed: the M5 Max (128 GB) buys more speed per dollar.
The $4,400 premium (86% more) buys +18% decode speed on Qwen 3.6-35B-A3B and 15 more models that fit.
Bottom line: The M5 Ultra (256 GB): a median 1.6× the speed for $4,400 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 Mac Studio M5 Ultra (256 GB), if the $4,400 premium fits: the LLMCheck index’s figures put it a median 1.6× faster across the 89 shared models with a figure on both; on Qwen 3.6-35B-A3B, 153 against 130 tok/s (both community runs). It also fits 15 more of the index’s models.
Decode speed on Apple Silicon follows memory bandwidth: the Mac Studio M5 Ultra (256 GB) reads memory at 1,228 GB/s, the Mac Studio M5 Max (128 GB) at 614 GB/s. The LLMCheck index puts Qwen 3.6-35B-A3B at 153 tok/s (community run) against 130 tok/s (community run). Across the 89 shared models with a figure on both, the median gain is 1.64x.
15 models in the LLMCheck index need more than 128 GB but fit in 256 GB, led by MiMo-V2.6-Flash, MiniMax-M3, GLM 5.2. Both run the other 93.
It costs $9,499 against $5,099 (Apple list prices, checked 14 Nov 2026; the M5 Ultra (256 GB) price includes 1TB of storage, the M5 Max (128 GB) price 512GB). For that it gives 153 tok/s (community run) against 130 tok/s (community run) on Qwen 3.6-35B-A3B and fits 15 more models. Worth it if you use the extra speed or models daily; the cheaper Mac still runs all but 15 of the same models.
On the Mac Studio M5 Max (128 GB): Qwen 3.6-35B-A3B at 130 tok/s (community run). On the Mac Studio M5 Ultra (256 GB): MiMo-V2.6-Flash at 77 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.