📚 More on this topic: Qwen 3.6 Setup Guide · Qwen 3.5 Setup Guide · DeepSeek V4 Flash vs Pro · Running LLMs on Mac M-Series · Qwen 3.5 on Mac: MLX vs Ollama · llama.cpp vs Ollama vs vLLM · VRAM Requirements · Run 31B Models on a Laptop
Every Mac with Apple Silicon can run local LLMs. The question isn’t whether — it’s which model, and whether it’ll be fast enough to actually use. A model that “fits” in memory but generates 3 tokens per second isn’t useful. A smaller model at 40 tok/s is.
This guide gives you specific model recommendations for every Mac tier, with real performance numbers. No “it depends” — concrete picks you can install right now. Updated May 2026 with Llama 4 Scout viability on 96GB+ Macs, MLX-community Qwen 3.6 still holding ~88K monthly downloads, and the ongoing M5 Max supply constraint context.
For setup instructions and general architecture details, see our complete M-series guide. This article focuses on which models to run.
What changed since the last refresh
The short version: the MoE revolution finally landed on Mac.
- Qwen 3.6-35B-A3B (released April 16, 2026) is a 35B MoE with only 3B active per token. That math changes everything on Mac — the model file is ~20GB at Q4, but token-generation speed feels like a 3B model. Native context 262K, extendable to ~1M via YaRN, Apache 2.0.
- Qwen 3.6-27B dense (released April 22, 2026) is the new flagship dense coding model. 77.2 on SWE-bench Verified. Apache 2.0. Simon Willison ran the 16.8GB Unsloth Q4_K_M GGUF and clocked 25.57 tok/s — flagship-class output on a single Mac.
- Gemma 4 26B-A4B landed with 4B active params and 256K context under Google’s Gemma license. Good alternative MoE if you want something other than Qwen.
- DeepSeek V4-Flash (284B total, 13B active, 1M context, MIT — see our V4 Flash vs Pro guide) is theoretically Mac-runnable at aggressive quant on 96GB+ configs, but no independent Mac benchmarks yet as of April 24.
- M5 Max MacBook Pro shipped in March 2026. Mac Studio M5 Max/Ultra is delayed until at least October 2026 due to memory shortages — Bloomberg’s Mark Gurman reported this on April 19. If you’re shopping a desktop, know that.
Qwen 3.5 is still fine. If you’re on Ollama and hit the qwen35moe mmproj bug or haven’t upgraded your tooling for 3.6 yet, Qwen 3.5-9B and 3.5-32B remain solid picks. But the defaults have moved.
Why Mac Is Different
Unified Memory Changes the Math
On a PC, your GPU has its own dedicated VRAM (typically 8-24GB). Models that don’t fit in VRAM either won’t run or crawl at 2-3 tok/s via offloading.
On Mac, there’s no separate GPU memory. Your entire RAM pool — 8GB to 192GB — is shared between CPU and GPU. A Mac Mini with 48GB can load a 32B model that would need a $700+ used RTX 3090 on PC. A Mac Studio with 128GB runs 70B models that require $3,000+ in dual GPUs.
The tradeoff: Mac’s memory bandwidth is lower than a discrete GPU’s. An RTX 3090 pushes 936 GB/s. The M4 Pro pushes 273 GB/s. Token generation speed is directly proportional to memory bandwidth, so Mac is 30-60% slower per token for models that fit in a GPU’s VRAM. But for models that don’t fit — Mac wins by running them at all.
MoE changes this equation. Qwen 3.6-35B-A3B only activates 3B params per token. That means token speed is closer to a 3B dense model’s, even though the file is 35B’s worth of disk and memory. This is why the new Qwen 3.6 MoE is such a big deal on Mac specifically.
Memory Bandwidth Matters More Than Chip Generation
This is the counterintuitive part. An M3 Max (400 GB/s) generates tokens faster than an M4 Pro (273 GB/s), despite the M4 Pro being a newer chip. For LLM inference, bandwidth is the bottleneck, not compute.
| Chip | Memory Bandwidth | Relative Speed |
|---|---|---|
| M1 / M2 / M3 / M4 / M5 (base) | 68-150 GB/s | 1x |
| M1 Pro / M2 Pro / M3 Pro / M4 Pro / M5 Pro | 150-307 GB/s | 2-2.5x |
| M1 Max / M2 Max / M3 Max / M4 Max / M5 Max | 300-614 GB/s | 3-5x |
| M1 Ultra / M2 Ultra / M3 Ultra | 400-800 GB/s | 4-7x |
Before buying: check the bandwidth of your specific chip, not just the generation. A Mac Mini M4 Pro 48GB is slower per token than a Mac Studio M4 Max 64GB, even on the same model. The M5 Max doubles the M4 Pro’s bandwidth — that’s the story for anyone shopping a laptop in 2026.
Best Models by Mac Tier
8GB Macs (M1 / M2 / M3 / M4 base)
macOS needs 2-3GB for itself. You have about 5-6GB for a model. This limits you to 3B-4B models comfortably or 7B-8B models with aggressive quantization and short context.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Gemma 4 E2B | ~1.5 GB | 30-45 tok/s | Google’s 2026 tiny model, strong at summarization |
| Qwen 3.5 4B | ~2.5 GB | 25-35 tok/s | Multilingual, good instruction following |
| Llama 3.2 3B | ~2 GB | 25-35 tok/s | General chat, still fine in 2026 |
| Phi-4 Mini 3.8B | ~2.3 GB | 25-40 tok/s | Reasoning-heavy tasks for its size |
| Qwen 3.5 9B Q3 | ~4 GB | 10-15 tok/s | Tight fit, quality tradeoff, short context only |
The pick: Gemma 4 E2B if your tool supports it, Qwen 3.5 4B as the safe fallback. Both fit easily, leave room for context, and stay fast.
Skip: Any 7B+ model at Q4 or higher. It’ll technically load but you’ll have 1-2GB for context and system, which means frequent crashes and 4K token limits.
Honest take: 8GB Macs are getting uncomfortable in 2026. 2026’s model generation — Qwen 3.6, Gemma 4 26B-A4B, DeepSeek V4 — is explicitly built for MoE configs with more memory headroom. If you’re serious about local AI, the memory upgrade is worth it. An M4 or M5 MacBook Air starts at 16GB now — that’s the minimum to buy going forward.
16GB Macs (M1 Pro 16GB / M2 16GB / M3 16GB / M4 16GB / M5 16GB)
The 7B-8B tier. You have ~12-13GB available for the model and context.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Qwen 3.5 9B Q4 | ~6.6 GB | 20-40 tok/s | Best all-rounder at this tier |
| Gemma 4 E4B | ~4.5 GB | 25-40 tok/s | Google’s efficient 4B variant, strong at chat |
| Llama 3.1 8B Q4 | ~4.5 GB | 25-40 tok/s | General assistant, well-tested |
| DeepSeek-R1-Distill-Qwen-8B | ~4.5 GB | 20-35 tok/s | Reasoning and chain-of-thought |
| Qwen 2.5 Coder 7B Q4 | ~4.5 GB | 25-40 tok/s | Previous-gen coding option (Qwen 3.5 9B preferred) |
The pick: Qwen 3.5 9B (Q4_K_M). This has held its slot since the last update — fits in ~6.6GB via Ollama, beats models 3x its size on reasoning, /think mode when you need chain-of-thought. See our 9B setup guide.
Worth testing: Gemma 4 E4B. Google’s smaller Gemma 4 variant has picked up steam on the Arena leaderboard. Worth a try if you want an alternative to Qwen.
Honest take: 16GB Macs are now the floor for useful local AI. You can run capable 8B-9B models, but the 32B+ tier with Qwen 3.6-35B-A3B is where 2026 gets interesting — and that requires 32GB+. If you’re buying new, step up.
24GB Macs (M2 Pro 24GB / M4 Pro 24GB / M4 16GB with swap)
The 14B tier and the low edge of the Qwen 3.6-27B zone.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Qwen 3.6-27B Q4_K_M | ~16.8 GB | 18-28 tok/s | Tight but doable — the new coding pick |
| Qwen 3 14B Q4 | ~9 GB | 15-30 tok/s | Safe general model at this tier |
| DeepSeek-R1-Distill-14B Q4 | ~8.5 GB | 15-25 tok/s | Complex reasoning, math, analysis |
| Mistral Nemo 12B Q4 | ~7.5 GB | 18-30 tok/s | 128K context for long documents |
| Qwen 3.5 9B Q8 | ~10 GB | 18-30 tok/s | Maximum quality at 9B size |
The pick: Qwen 3.6-27B (Unsloth Q4_K_M GGUF). Per Simon Willison’s April 22 post, this runs at 25.57 tok/s on his machine with the 16.8GB GGUF — flagship-class coding output on a single Mac. On a 24GB Mac you’re at the knife’s edge: model is 16.8GB, leaving ~5GB for macOS and context. Expect shorter context limits and occasional memory pressure. Good tradeoff if coding is the workload.
Safer pick: Qwen 3 14B. If you don’t want to live on the edge, 14B at Q4 leaves comfortable context headroom and still delivers strong general output.
Don’t bother with: Qwen 3 32B at Q3. It technically fits but quality at Q3 is degraded enough that a 14B at Q4 usually wins. Wait for more memory.
32-48GB Macs (M3 Pro 36GB / M4 Pro 48GB / M2-M4 Max 32-48GB)
This is where Qwen 3.6-35B-A3B MoE wins outright. The model fits, the active-param footprint is small, and you get room for context.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Qwen 3.6-35B-A3B Q4 (MoE) | ~20 GB | 35-55 tok/s | The 2026 default — fast MoE, strong all around |
| Qwen 3.6-27B Q4_K_M (dense) | ~16.8 GB | 18-28 tok/s | Best coding model under 50GB |
| Gemma 4 26B-A4B Q4 (MoE) | ~15 GB | 30-45 tok/s | Alternative MoE, 256K context |
| Gemma 4 31B-it Q4 (dense) | ~17 GB | 18-28 tok/s | Dense Gemma 4 alternative — sliding-window attention |
| Qwen 3 32B Q4 (dense) | ~20 GB | 12-22 tok/s | Previous-gen pick, still fine |
| DeepSeek-R1-Distill-32B Q4 | ~20 GB | 12-22 tok/s | Reasoning, math, complex analysis |
| Qwen 2.5 Coder 32B Q4 | ~20 GB | 12-22 tok/s | Previous-gen coding pick |
The pick: Qwen 3.6-35B-A3B (Q4_K_M via Ollama, MLX 4-bit via MLX-LM). This is the model that makes Mac local AI worthwhile in 2026. MoE design means 3B active params per token, so token speed stays high even as the 20GB file sits in memory. Native 262K context. Strong on code (73.4 SWE-bench Verified), strong on reasoning (92.7 AIME26), strong on general tasks (85.2 MMLU-Pro).
MLX variants are live at mlx-community/Qwen3.6-35B-A3B-4bit and unsloth/Qwen3.6-35B-A3B-UD-MLX-4bit.
For coding specifically: Qwen 3.6-27B dense. Higher single-token quality than the A3B MoE on code. Slower per token (dense model hits full 27B per step), but the output is worth it. Simon Willison’s verdict on the Q4 GGUF: “an outstanding result for a 16.8GB local model.”
Alternative MoE: Gemma 4 26B-A4B. 4B active, 256K context, different training recipe. Worth having on disk for variety.
The Mac Mini M4 Pro 48GB at $1,799 remains the best-value setup in this tier. Silent, low power, runs Qwen 3.6-35B-A3B all day with room for 32K+ context.
48-64GB Macs (M3 Max 48-64GB / M4 Max 64GB / M5 Max 48-64GB)
70B models become practical. Qwen 3.6-35B-A3B runs with generous context. You can keep multiple models resident.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Qwen 3.6-35B-A3B Q8 (MoE) | ~37 GB | 25-45 tok/s | Near-full quality MoE, fast |
| Qwen 3.6-27B Q6/Q8 (dense) | ~22-30 GB | 12-22 tok/s | Best local coding model, higher quant |
| Llama 3.3 70B Q4 | ~40 GB | 8-15 tok/s | General-purpose large model |
| Qwen 2.5 72B Q4 | ~42 GB | 8-14 tok/s | Previous-gen multilingual pick (strong CN/JP/KR) |
| DeepSeek-R1-Distill-70B Q4 | ~40 GB | 8-14 tok/s | Reasoning at scale |
| Gemma 4 26B-A4B Q8 | ~28 GB | 25-40 tok/s | Higher-quality Gemma MoE |
The pick: Qwen 3.6-35B-A3B at Q8. At this memory tier you can afford the higher quant. Quality improvement over Q4 is meaningful on edge cases, and the MoE architecture keeps it fast.
For coding: Qwen 3.6-27B at Q6 or Q8. Pairs the best dense coding model with high-quality quantization.
If you want a 70B: Llama 3.3 70B is still the standard at Q4. 8-15 tok/s on M4 Max — slower than reading speed but usable for interactive work.
96-128GB Macs (M2/M3 Ultra, M4 Max 128GB)
The “no compromises” tier for consumer hardware. Frontier-class models become practical.
| Model | Size | Speed | Best For |
|---|---|---|---|
| Qwen 3.6-35B-A3B bf16 (full precision) | ~70 GB | 20-35 tok/s | Maximum quality on the MoE |
| Llama 4 Scout (MoE, 17B active) | ~58 GB (Q4) | 25-35 tok/s on M5 Max | 109B MoE in reach, native 10M context |
| DeepSeek V4-Flash Q4 (MoE) | ~140 GB | Unknown on Mac | Aspirational — test if you have it |
| Llama 3.1 70B Q6 | ~55 GB | 8-15 tok/s | Maximum 70B quality |
| Qwen 2.5 72B Q8 | ~75 GB | 8-12 tok/s | Previous-gen 72B at near-lossless quant |
| Qwen3 235B-A22B Q4 | ~88 GB | 5-10 tok/s | Larger Qwen MoE, previous-gen |
The pick: Qwen 3.6-35B-A3B at bf16 if you want to see what full-precision Qwen 3.6 looks like on a consumer Mac. This is as close to the Hugging Face Inference API experience as you’ll get locally.
DeepSeek V4-Flash — worth testing. The 284B/13B-active MoE is the aspirational model for this tier. At aggressive MLX 4-bit quantization it’s plausible on 128GB — theoretical footprint around 140GB for weights plus KV cache for whatever context you load, which means 128GB is tight and 192GB is comfortable. No independent Mac benchmarks yet as of April 24, 2026. If you have the hardware, pull the weights and report back. See our V4 Flash vs Pro guide for current sourcing notes.
Llama 4 Scout is the practical MoE pick at this tier. 109B total / 17B active fits in ~58 GB at Q4, leaving room for context on 96GB+ Macs. M5 Max users have reported ~32 tok/s via MLX. MoE caveat: Scout’s expert-routing degrades more on aggressive quant than dense models — the router operates on weight distributions, and Q4 loses enough precision to pick wrong experts more often. If memory allows, run Q5 instead of Q4. At 96GB you have the headroom for it; at 128GB it’s a no-brainer.
Skip DeepSeek V4-Pro. 1.6T total / 49B active. Even at aggressive quant it’s ~800GB of weights. Not a Mac story, period.
192GB+ Macs (M3 Ultra 192GB, rare M2 Ultra top config)
The rarefied tier. Research-grade local AI.
| Model | Size | Speed | Best For |
|---|---|---|---|
| DeepSeek V4-Flash Q4-Q6 | ~140-180 GB | No Mac benchmarks yet | Frontier-adjacent MoE |
| Qwen3 235B-A22B Q6-Q8 | ~140-180 GB | 4-8 tok/s | Largest Qwen MoE at higher quant |
| Llama 3.1 405B Q3 | ~150 GB | 2-4 tok/s | Frontier dense model, slow |
The pick: DeepSeek V4-Flash is the most interesting model to try at this tier. 1M native context window, MIT licensed, and the 13B active-param footprint means token speed should stay reasonable once weights are loaded. MLX support for the V4 architecture is pending at time of writing — watch the mlx-community HF org.
If you have a 192GB Ultra and you’re not experimenting with 200B+ parameter models, what are you even doing with it.
MLX vs Ollama vs LM Studio in 2026
All three work on Apple Silicon. The difference is speed, ease, and which models are supported on day one.
| Tool | Backend | Speed (Qwen 3.6-35B-A3B Q4, M4 Max) | Setup | Best For |
|---|---|---|---|---|
| MLX-LM | Apple MLX | ~45-55 tok/s (MoE) | Python CLI | Maximum speed when supported |
| Ollama | llama.cpp | ~35-45 tok/s (MoE) | One command | Simplest setup, API server |
| LM Studio | llama.cpp + MLX | ~40-55 tok/s (MoE) | GUI app | Visual interface, model browsing |
Numbers above are approximate community reports on Qwen 3.6-35B-A3B Q4 on M4 Max configs. Individual mileage varies substantially by RAM size, concurrent load, and quant variant.
When to Use MLX
MLX is Apple’s native machine learning framework, built for unified memory. On supported models it’s typically 10-30% faster than llama.cpp. The gap used to be larger — llama.cpp has closed a lot of ground on Mac since 2025, and for some recent models at Q4 GGUF the gap is near-zero.
Use MLX when:
- You want maximum tok/s and you’re comfortable with Python
- You’re building apps that need fast local inference
- You’re on Qwen 3.6 or Gemma 4 specifically — MLX Community has full 3.6 coverage now
pip install mlx-lm
mlx_lm.generate --model mlx-community/Qwen3.6-35B-A3B-4bit --prompt "Hello"
The mlx-community HF org maintains MLX-converted variants of nearly every mainstream model within days of release. Unsloth also ships MLX builds now — see unsloth/Qwen3.6-35B-A3B-UD-MLX-4bit. For a head-to-head speed comparison on the Qwen 3.5 family specifically, see our MLX vs Ollama benchmark on Apple Silicon.
When to Use Ollama
Ollama wraps llama.cpp with dead-simple model management. One command to install, one command to run.
Use Ollama when:
- You want the fastest setup possible
- You need an API server for other apps (Open WebUI, Continue, etc.)
- You’re new to local LLMs
- You want to switch between models quickly
ollama run qwen3.6:35b-a3b-q4_K_M
Known Ollama gotcha: the qwen35moe mmproj bug, filed March 9, 2026 and closed as a duplicate of #14575, affects Qwen 3.5 MoE variants when used with separate vision projector files. Ollama errors with “unknown model architecture: ‘qwen35moe’”. Workaround: use llama.cpp directly with --mmproj, or switch to Qwen 3.6 for vision work (different arch, not affected by this specific bug). Monitor the tracking issue before committing to a vision pipeline on Qwen 3.5 MoE via Ollama.
When to Use LM Studio
LM Studio gives you a ChatGPT-like interface with model browsing, parameter controls, and conversation management. Recent versions use MLX as the backend on Mac when available, closing the speed gap with MLX-LM.
This is the tool Simon Willison used to run Qwen 3.6-35B-A3B on his MacBook Pro M5 with the UD-Q4_K_S 20.9GB GGUF. His verdict after the pelican-on-a-bicycle test: “I’m giving this one to Qwen 3.6. Opus managed to mess up the bicycle frame!” — a light result, but one he didn’t expect.
Use LM Studio when:
- You prefer a GUI over terminal
- You want to browse and compare models visually
- You need fine control over temperature, top-p, and sampling
- You want MLX backend without writing Python
DeepSeek V4 on Mac: the honest take
DeepSeek V4 dropped the evening of April 23, 2026 — two variants, MIT license, 1M context. See our V4 Flash vs Pro guide for the full breakdown. For Mac specifically:
V4-Pro (1.6T / 49B active). Not a Mac story. Even on a 192GB M3 Ultra, the weights don’t fit without disk offload, and disk offload at 1M context is not going to be fun. Server hardware only.
V4-Flash (284B / 13B active). Theoretically Mac-runnable at MLX 4-bit on 128GB+ configs. The math: weights at aggressive quant are ~140GB, plus KV cache for context. 128GB is tight; 192GB is comfortable. No independent Mac benchmarks yet as of April 24 — if you pull it and run it, you’re one of the first. MLX community support is pending; llama.cpp support via GGUF is not yet published by the usual quantizers (Unsloth, bartowski). Expect that to land within a week or two of release.
What to actually do: if you’re on 96GB+ and curious, monitor the mlx-community HF org for V4-Flash MLX uploads. If the MLX variant doesn’t arrive within 10 days, wait for llama.cpp GGUF support and Unsloth quants. Don’t burn time converting yourself.
M5 Max and the Mac Studio Situation
M5 Max MacBook Pro shipped in March 2026. Bandwidth is up substantially from M4 Max — the M5 Max / M5 Pro guide covers the numbers. For local LLM work specifically, the bandwidth bump matters more than the compute bump.
The complication: Mac Studio M5 Max and M5 Ultra are delayed until at least October 2026. Bloomberg’s Mark Gurman reported on April 19, 2026 that industry-wide memory and storage shortages driven by AI demand pushed the Studio refresh back. Apple is reportedly prioritizing MacBook shipments over desktops during the constrained period.
What this means practically:
- Need a desktop for local AI in the next 6 months? Buy a Mac Studio M4 Max or M3 Ultra now. Don’t wait for M5.
- Need a portable AI workstation? M5 Max MacBook Pro is the move.
- Want M5 Ultra specifically? October 2026 earliest, probably later.
Don’t trust leak timelines for new Apple Silicon Ultra variants. The M3 Ultra Mac Studio was a surprise release. Plan around what’s shipping, not what’s rumored.
Models That Technically Fit But Actually Crawl
This is the trap. A model can load into memory and still be useless.
| Scenario | What Happens | Speed |
|---|---|---|
| 70B Q4 on 64GB M4 Pro | Model loads, but only 4GB for context. M4 Pro’s 273 GB/s bandwidth makes it slow. | 4-7 tok/s |
| Qwen 3.6-27B Q4 on 24GB | Model is 16.8GB, leaving ~5GB for OS and context. Tight. | 15-25 tok/s with context pressure |
| DeepSeek V4-Flash Q4 on 96GB | Weights are ~140GB. Won’t fit. | Will not run |
| Mixtral 8x7B on 32GB | All 46.7B params load (~26GB Q4). Runs but barely any context headroom. | 8-12 tok/s, 4K context max |
| Qwen3 235B-A22B on 64GB | Weights are ~88GB Q4. Won’t fit without disk offload. | Will not run comfortably |
The rule of thumb: the model file should be no more than 60-70% of your total memory. That leaves room for macOS, the KV cache (context), and framework overhead. A 20GB model on a 48GB Mac is comfortable. A 20GB model on a 24GB Mac is a knife’s edge.
If you’re right at the limit, drop to a lower quantization or a smaller model. A snappy 14B model is more useful than a sluggish 32B.
Want to check the math for your specific Mac? Use the Local AI Planning Tool — pick your Mac’s unified memory, see every model that fits at your quant, with per-model VRAM breakdowns including KV cache at your target context length.
The Best Mac for Local AI in 2026
| Budget | Buy This | Best Model It Runs | Why |
|---|---|---|---|
| $599 | Mac Mini M4 16GB | Qwen 3.5 9B | Cheapest usable entry point |
| $1,399 | Mac Mini M4 Pro 24GB | Qwen 3.6-27B Q4 (tight) | Decent coding setup |
| $1,799 | Mac Mini M4 Pro 48GB | Qwen 3.6-35B-A3B | Best value — the 2026 default |
| $2,700 | Mac Studio M4 Max 64GB | Qwen 3.6-35B-A3B Q8 + Llama 3.3 70B | Fastest bandwidth, keep multiple models |
| $3,500 | Mac Studio M4 Max 128GB | DeepSeek V4-Flash (aspirational) | No compromises on current gen |
| $5,500+ | Mac Studio M3 Ultra 192GB | DeepSeek V4-Flash comfortably | Research-grade local AI |
The Mac Mini M4 Pro 48GB at $1,799 is still the sweet spot. It runs Qwen 3.6-35B-A3B comfortably, sits silently on your desk, draws 30W under AI load, and costs less per year in electricity than a single month of ChatGPT Plus.
If you’re shopping a laptop, M5 Max MacBook Pro 64GB is the best answer for portable local AI in 2026.
The Bottom Line
The model matters more than the tool. Pick the right model for your memory tier, use whichever app you’re comfortable with, and don’t try to squeeze a model that’s too big. A fast small model beats a slow big one every time.
Quick decision tree for 2026:
- 8GB: Gemma 4 E2B or Qwen 3.5 4B. Accept the limitations. The platform is outgrowing you.
- 16GB: Qwen 3.5 9B. Still useful, but the interesting 2026 models need more memory.
- 24GB: Qwen 3.6-27B Q4 if you can tolerate tight context; Qwen 3 14B if you want comfort.
- 32-48GB: Qwen 3.6-35B-A3B is the 2026 default. Fast MoE, strong all around, fits comfortably with 32K+ context.
- 48GB+: Qwen 3.6-27B dense at Q6/Q8 is the best local coding model. Test Simon Willison’s setup.
- 64-96GB: Qwen 3.6-35B-A3B at Q8 plus a 70B resident. Gemma 4 26B-A4B for variety. Llama 4 Scout becomes plausible at 96GB.
- 128GB+: DeepSeek V4-Flash becomes plausible. Worth testing. No consumer-hardware benchmarks yet, so you’re early.
- 192GB Ultra: DeepSeek V4-Flash comfortably. You’ve won the local AI hardware lottery.
Install Ollama (curl -fsSL https://ollama.com/install.sh | sh) or MLX-LM (pip install mlx-lm), pull your model, and start chatting. Your Mac is already a capable AI workstation — you just need to pick the right model for it.
The defaults moved in April 2026. If you haven’t pulled Qwen 3.6 yet, that’s the weekend project.
Last updated May 24, 2026. Added Llama 4 Scout consideration for 96GB+ Macs, Gemma 4 31B-it dense alternative, tightened Qwen 2.5 legacy labels, M5 Max supply constraint context updated.
Get notified when we publish new guides.
Subscribe — free, no spam