Best Local LLMs for the MacBook Pro M3 Max (128 GB)

The best local LLM for a MacBook Pro M3 Max (128 GB) is Qwen3.8-Flash-Next at ~35 tok/s (estimated). Its 128 GB fits 93 of the 130 models in the LLMCheck index, up to 320B-class. For everyday chat and coding, Qwen3.8-Flash-Next is the sweet spot.

Unified memory
128
GB
Mem. bandwidth
400
GB/s
Models that fit
93
of 130
Top speed
estimated255
tok/s

Considering this Mac?
MacBook Pro M5 Max (128GB): build-to-order from Apple · MacBook Pro M3 Max on Amazon Renewed → · compare every Mac for local AI →

Top 3 picks for the MacBook Pro M3 Max (128 GB)

⭐ Best overall
125B MoE · Qwen Custom · cap 47/50 raise GPU limit
Loads only with the GPU limit raised past the index’s safe setting (measured runs on this much memory did): see the GPU memory limit note
estimated35 tok/s
⚡ Fastest
1.73B · LFM Open · cap 17/50 · agent-ready 64K
estimated255 tok/s
🧠 Most capable
320B MoE · MIT · cap 48/50 over GPU limit
estimated17 tok/s

Every model ranked for a MacBook Pro M3 Max (128 GB)

Ranked by LLMCheck suitability (capability balanced against speed on the M3 Max); 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 →

#ModelSizeLicenseSpeed64K contextCapability
1Qwen3.8-Flash-Next raise GPU limit125B MoEQwen Customestimated35 tok/sneeds ~122 GB47/50
2DeepSeek V4 Flash raise GPU limit284B MoEMITcommunity27 tok/sneeds ~121 GB47/50
3GLM-5.3-Flash over GPU limit320B MoEMITestimated17 tok/sneeds ~129 GB48/50
4DeepSeek V4 Flash Vision-Exp over GPU limit305B MoEMITestimated24 tok/sneeds ~121 GB47/50
5Inkling-Small276B MoEApache 2.0estimated29 tok/sagent-ready46/50
6KAT-Coder-V2.535B MoEApache 2.0estimated84 tok/sagent-ready39/50
7Qwen 3.6-35B-A3B35B MoEApache 2.0estimated92 tok/sagent-ready38/50
8Qwen3.8-27B27.8BApache 2.0estimated19 tok/sagent-ready46/50
9Qwen3-235B-A22B over GPU limit235B MoEApache 2.0estimated17 tok/sneeds ~140 GB46/50
10Nanbeige4.3-33B33.1BApache 2.0estimated15 tok/sagent-ready46/50
11Recursive Flint39.8B MoEApache 2.0estimated89 tok/sagent-ready37/50
12Ling-3.0-flash-VL124B MoEMITestimated37 tok/sagent-ready42/50
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Showing the top 12 of 93 models that fit in 128 GB. See the full leaderboard or all benchmarks.

Estimated figures come from the bandwidth formula; against 233 measured runs its typical error is ±9% for dense models, ±15% for small MoEs (≤40B total) and ±34% for large MoEs (mostly below the measured figure). How the error is computed →

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 112 GB this Mac’s GPU can use with the limit raised (76 of the 85 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 96 GB of this Mac’s 128 GB by default on macOS 15 and earlier, and about 107.5 GB on macOS 26 and later. Of the 93 models that fit, 7 need more than that at a short prompt: 2 need a raised limit and are tagged raise GPU limit (Qwen3.8-Flash-Next and DeepSeek V4 Flash only past the safe setting below, as measured runs on this much memory show); 5 need more than even the raised limit and are tagged over GPU limit. sudo sysctl iogpu.wired_limit_mb=114688 raises it to 112 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 →

Quick start: run Qwen3.8-Flash-Next on your MacBook Pro M3 Max

GPU memory limit: Qwen3.8-Flash-Next needs about 120 GB, more than macOS lets the GPU use on this Mac by default (about 96 GB). Measured runs on 128 GB Macs show it loads with the limit raised, past the index’s safe setting (sudo sysctl iogpu.wired_limit_mb=114688, 112 GB), so macOS keeps less than 16 GB: quit other apps first.

The index has no verified Ollama tag for Qwen3.8-Flash-Next, so start from its listing and copy the exact name there: Search the Ollama library → · 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 Qwen3.8-Flash-Next on M3 Max page for its figures and setup.

Storage

The models a 128 GB Mac can hold are big files: Qwen3-235B-A22B, the largest that fits here, is about a 126 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.

Same Mac with more memory, and the upgrade path

Upgrade or save: Qwen3.8-Flash-Next on other Macs

Same pick, other Macs

Twice as fast: no Mac in the index runs Qwen3.8-Flash-Next at twice this Mac’s estimated~35 tok/s; the fastest figure is the Mac Studio M1 Ultra (128 GB) at estimated~57 tok/s, about $2,400 refurbished or used: search Amazon Renewed listings →.

For less: the cheapest Mac that still runs Qwen3.8-Flash-Next at 15+ tok/s is the Mac Studio M4 Max (128 GB) at estimated~45 tok/s, about $2,399 used (no longer sold new): 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.

🛒 The current Mac with the same 128 GB

Apple no longer sells the MacBook Pro M3 Max (128 GB) new. A current MacBook Pro M5 Max (128 GB) runs the same models with 614 GB/s of memory bandwidth instead of 400 GB/s, so they generate faster. The MacBook Pro M5 Max (128 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.

Cite this page

LLMCheck. (2026). Best Local LLMs for the MacBook Pro M3 Max (128 GB). llmcheck.net. https://llmcheck.net/best-llm/macbook-pro-m3-max-128gb/ (updated ). CC BY 4.0.

How to cite LLMCheck →Data: CC BY 4.0Markdown version

FAQ: local LLMs on the MacBook Pro M3 Max

What is the best local LLM for a MacBook Pro M3 Max (128 GB)?

Qwen3.8-Flash-Next (125B MoE, Qwen Custom) is the best all-round pick at ~35 tok/s (estimated) on the MacBook Pro M3 Max. If you want maximum speed, LFM3-1.7B hits ~255 tok/s (estimated); for maximum capability, GLM-5.3-Flash still fits in 128 GB.

How many models can a MacBook Pro M3 Max with 128 GB run?

About 93 of the 130 models in the LLMCheck index fit in 128 GB of unified memory, from compact models up to GLM-5.3-Flash (320B MoE).

Can a MacBook Pro M3 Max run a 70B model?

Yes. A dense 70B such as Llama 3.3 70B needs about 44 GB at 4-bit, which fits in 128 GB with room for context; the index puts it at ~8 tok/s (estimated) on this Mac.

Is 128 GB of RAM enough to run LLMs locally?

128 GB is enough for dense models up to about 128B at 4-bit (the largest in the index that fits is Mistral Medium 3.5), and for mixture-of-experts models such as GLM-5.3-Flash (~17 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.

Related