The best local LLM for a Mac Studio M3 Ultra (256 GB) is MiMo-V2.6-Flash at 59 tok/s (community run). Its 256 GB fits 108 of the 130 models in the LLMCheck index, up to 753B-class. For everyday chat and coding, MiMo-V2.6-Flash is the sweet spot.
Considering this Mac?
Mac Studio M5 Ultra (256GB): build-to-order from Apple · Mac Studio M3 Ultra on Amazon Renewed → · compare every Mac for local AI →
Ranked by LLMCheck suitability (capability balanced against speed on the M3 Ultra); the top pick is the best-ranked model that reaches 15 tok/s here, has a standard Mac runtime and fits the GPU memory limit (raised if need be), and slower models are tagged slow. Click a model for its full figures and setup. Every speed carries its provenance mark: most are index estimates from the memory-bandwidth model, a filled square is a published figure and a hollow circle a community run — submit a real run →
| # | Model | Size | License | Speed | 64K context | Capability |
|---|---|---|---|---|---|---|
| 1 | MiMo-V2.6-Flash | 309B MoE | MIT | community59 tok/s | agent-ready | 49/50 |
| 2 | Qwen3.8-Flash-Next | 125B MoE | Qwen Custom | estimated57 tok/s | agent-ready | 47/50 |
| 3 | Inkling 2 Small raise GPU limit | 594B MoE | Apache 2.0 | estimated36 tok/s | agent-ready | 48/50 |
| 4 | GLM 5.2 raise GPU limit | 753B MoE | MIT | estimated15 tok/s | 50/50 | |
| 5 | MiniMax-M3 | 428B MoE | MiniMax Community | estimated29 tok/s | agent-ready | 48/50 |
| 6 | GLM-5.3 slow over GPU limit | 753B MoE | GLM-5.3 | estimated11 tok/s | 50/50 | |
| 7 | MiniMax-M3.1-Flash | 428B MoE | MiniMax Community | community26 tok/s | agent-ready | 48/50 |
| 8 | KAT-Coder-V2.5 | 35B MoE | Apache 2.0 | estimated100 tok/s | agent-ready | 39/50 |
| 9 | Nemotron 3.5 Lightning | 30B MoE | OpenMDW | estimated151 tok/s | agent-ready | 36/50 |
| 10 | North Mini Code 1.0 | 30.5B MoE | Apache 2.0 | estimated133 tok/s | agent-ready | 36/50 |
| 11 | GLM-5.3-Flash | 320B MoE | MIT | estimated24 tok/s | agent-ready | 48/50 |
| 12 | DeepSeek V4 Flash | 284B MoE | MIT | estimated32 tok/s | agent-ready | 47/50 |
Showing the top 12 of 108 models that fit in 256 GB. See the full leaderboard or all benchmarks.
Context: “fits” means the weights at a short prompt. Coding agents (Claude Code, OpenCode, Codex CLI) hold 64K tokens or more, and every token adds KV cache: agent-ready marks a model whose weights plus a 64K-token 16-bit KV cache still fit in the 240 GB this Mac’s GPU can use with the limit raised (94 of the 98 models here with a KV figure). That is an estimate from each model’s config.json, not a measurement; a q8 KV cache halves the cache part, and — means the index has no KV figure for the model yet. The GPU memory limit below applies on top.
GPU memory limit: macOS lets the GPU use about 192 GB of this Mac’s 256 GB by default on macOS 15 and earlier, and about 222.7 GB on macOS 26 and later. Of the 108 models that fit, 8 need more than that at a short prompt: 7 need a raised limit and are tagged raise GPU limit (GLM 5.2 only past the safe setting below, as measured runs on this much memory show); 1 needs more than even the raised limit and is tagged over GPU limit. sudo sysctl iogpu.wired_limit_mb=245760 raises it to 240 GB, leaving 16 GB for macOS, until the Mac restarts (sudo sysctl iogpu.wired_limit_mb=0 restores the default). How the limit is read →
The index has no verified Ollama tag for MiMo-V2.6-Flash, so start from its listing and copy the exact name there: Hugging Face repo → · model card →
GGUF builds run in LM Studio or llama.cpp; MLX builds run in LM Studio or mlx-lm.
Prefer a GUI? LM Studio gives you a one-click download and chat window. For step-by-step help see our Ollama install guide, or open the MiMo-V2.6-Flash on M3 Ultra page for its figures and setup.
The models a 256 GB Mac can hold are big files: GLM-5.3, the largest that fits here, is about a 254 GB download at the build it needs, and a few models that size fill an internal drive. A fast external Thunderbolt 5 or USB4 SSD keeps a model library within reach: Thunderbolt 5 external SSDs on Amazon →
Amazon affiliate link: as an Amazon Associate, LLMCheck earns from qualifying purchases.
Twice as fast: no Mac in the index runs MiMo-V2.6-Flash at twice this Mac’s community59 tok/s; the fastest figure is the Mac Studio M5 Ultra (256 GB) at community77 tok/s, $9,499 new (Apple list price, checked 2026-11-14): build-to-order from Apple (Amazon does not stock this configuration).
For less: the cheapest Mac that still runs MiMo-V2.6-Flash at 15+ tok/s is the Mac Studio M2 Ultra (192 GB) at estimated~54 tok/s, about $3,000 refurbished or used: search Amazon Renewed listings →.
Speeds are each Mac’s own figure from the index, with its provenance mark. Prices come from data/macs.json: Apple’s list price for a new Mac (dated), an approximate street price for a used or refurbished one. Amazon stock checked 2026-11-18. Amazon links are affiliate links: as an Amazon Associate, LLMCheck earns from qualifying purchases.
Apple no longer sells the Mac Studio M3 Ultra (256 GB) new. A current Mac Studio M5 Ultra (256 GB) runs the same models with 1228 GB/s of memory bandwidth instead of 819 GB/s, so they generate faster. The Mac Studio M5 Ultra (256 GB) is build-to-order from Apple; Amazon does not stock that configuration.
As an Amazon Associate, LLMCheck earns from qualifying purchases. Affiliate links cost you nothing extra and never influence our rankings.
LLMCheck. (2026). Best Local LLMs for the Mac Studio M3 Ultra (256 GB). llmcheck.net. https://llmcheck.net/best-llm/mac-studio-m3-ultra-256gb/ (updated ). CC BY 4.0.
MiMo-V2.6-Flash (309B MoE, MIT) is the best all-round pick at 59 tok/s (community run) on the Mac Studio M3 Ultra. If you want maximum speed, LFM3-1.7B hits ~334 tok/s (estimated); for maximum capability, GLM 5.2 still fits in 256 GB.
About 108 of the 130 models in the LLMCheck index fit in 256 GB of unified memory, from compact models up to GLM 5.2 (753B MoE).
Yes. A dense 70B such as Llama 3.3 70B needs about 44 GB at 4-bit, which fits in 256 GB with room for context; the index puts it at ~16 tok/s (estimated) on this Mac.
256 GB is enough for dense models up to about 405B at 4-bit (the largest in the index that fits is Llama 3.1 405B), and for mixture-of-experts models such as GLM 5.2 (~15 tok/s (estimated)). As a rule of thumb a 4-bit model needs about 0.57 GB per billion parameters plus ~2 GB, and macOS needs its own share of memory too. Because Apple Silicon uses unified memory, the same pool serves as system RAM and GPU memory.