The verdict is the Mac Studio M5 Ultra (256 GB): the LLMCheck index gives both Macs the same speed on every model they share, and it runs the same headline pick, MiMo-V2.6-Flash, at 77 tok/s (community run). The $12,800 extra for the M5 Ultra (512 GB) buys 14 more models, led by DeepSeek V4.1 Flash.
| M5 Ultra (256 GB) | M5 Ultra (512 GB) | |
|---|---|---|
| Chip | M5 Ultra | M5 Ultra |
| GPU cores | 64- or 80-core | 80-core |
| Memory bandwidth | 1,228 GB/s | 1,228 GB/s |
| Unified memory | 256 GB | 512 GB |
| Price | $9,499 | $22,299 |
| Price per GB | $37 | $44 |
| Models that fit (of 130) | 108 | 122 |
| Best local LLM here | MiMo-V2.6-Flash community77 tok/s | MiMo-V2.6-Flash community77 tok/s |
| MiMo-V2.6-Flash | community77 tok/s | community77 tok/s |
| Price per tok/s on MiMo-V2.6-Flash | $123 | $290 |
| 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 Ultra (256 GB) and Mac Studio M5 Ultra (512 GB) pages print, with its provenance mark. Across the 104 shared models with a figure on both, the 512 GB column is a median ×1.00 of the 256 GB column.
| Model | Size | 256 GB tok/s | 512 GB tok/s | Ratio |
|---|---|---|---|---|
| MiMo-V2.6-Flash | 309B MoE | community77 | community77 | ×1.00 |
| Qwen3.8-Flash-Next | 125B MoE | community57 | community57 | ×1.00 |
| KAT-Coder-V2.5 | 35B MoE | estimated120 | estimated120 | ×1.00 |
| MiniMax-M3 | 428B MoE | estimated42 | estimated42 | ×1.00 |
| Qwen 3.6-35B-A3B | 35B MoE | community153 | community153 | ×1.00 |
| Ling-3.0-flash-VL | 124B MoE | estimated89 | estimated89 | ×1.00 |
| GLM 5.2 | 753B MoE | estimated22 | estimated22 | ×1.00 |
| DeepSeek V4 Flash Vision-Exp | 305B MoE | estimated46 | estimated46 | ×1.00 |
| DeepSeek V4 Flash | 284B MoE | estimated45 | estimated45 | ×1.00 |
| Inkling 2 Small caveat | 594B MoE | estimated36 | estimated36 | ×1.00 |
| Nanbeige4.3-33B | 33.1B | estimated52 | estimated52 | ×1.00 |
| GLM-5.3-Flash | 320B MoE | estimated35 | estimated35 | ×1.00 |
| MiniMax-M3.1-Flash | 428B MoE | community35 | community35 | ×1.00 |
| GLM-5.3 | 753B MoE | estimated17 | estimated17 | ×1.00 |
| Recursive Flint | 39.8B MoE | estimated131 | estimated131 | ×1.00 |
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 | 512 GB tok/s |
|---|---|---|---|
| DeepSeek V4.1 Flash | 552B MoE | 434 GB | estimated29 |
| Muse Spark Open | 816B MoE | 342 GB | estimated31 |
| Kimi K2.6 | 1T MoE | 429 GB | estimated27 |
| Kimi K2.7 Code | 1T MoE | 421 GB | estimated27 |
| Kimi K2.5 | 1T MoE | 417 GB | estimated28 |
| Ling-3.1-flash | 560B MoE | 360 GB | estimated27 |
| GLM-5.1 | 754B MoE | 390 GB | estimated17 |
| Hy4 preview caveat | 770B MoE | 292 GB | estimated14 |
| Mistral Large 4 | 1.05T MoE | 383 GB | estimated21 |
| DeepSeek V3.2-Speciale | 685B MoE | 360 GB | estimated18 |
At Apple’s list prices the M5 Ultra (256 GB) costs $37 per GB of unified memory and the M5 Ultra (512 GB) $44. On MiMo-V2.6-Flash — the reference model here, which both run — that is $123 and $290 per tok/s of decode speed: the M5 Ultra (256 GB) buys more speed per dollar.
The $12,800 premium (135% more) buys no extra decode speed on MiMo-V2.6-Flash and 14 more models that fit.
Bottom line: The M5 Ultra (256 GB): same speeds; the extra memory buys 14 more models.
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): the LLMCheck index gives both Macs the same speed on every model they share, and it runs the same headline pick, MiMo-V2.6-Flash, at 77 tok/s (community run). The $12,800 extra for the M5 Ultra (512 GB) buys 14 more models, led by DeepSeek V4.1 Flash.
No. Both use the M5 Ultra at 1,228 GB/s, and decode speed on Apple Silicon follows memory bandwidth, so the LLMCheck index gives the same figure on every one of the 104 shared models (MiMo-V2.6-Flash: 77 tok/s (community run)). The extra 256 GB buys models and context, not speed.
14 models in the LLMCheck index need more than 256 GB but fit in 512 GB, led by DeepSeek V4.1 Flash, Muse Spark Open, Kimi K2.6. Both run the other 108.
It costs $22,299 against $9,499 (Apple list prices, checked 14 Nov 2026). For that it gives the same 77 tok/s (community run) on MiMo-V2.6-Flash and fits 14 more models. For local LLMs the cheaper Mac is the better buy; pay more for the extra 256 GB of headroom, not for speed.
On the Mac Studio M5 Ultra (256 GB): MiMo-V2.6-Flash at 77 tok/s (community run). On the Mac Studio M5 Ultra (512 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.