Not sure which model fits your Mac?
Use our free checker — select your hardware and get instant recommendations.
Nemotron 3.5 Bolt E4B on a Mac: NVIDIA’s “4 GB” Model Needs 6 GB
NVIDIA's Nemotron 3.5 Bolt E4B runs on Macs in Ollama, llama.cpp and MLX. Its '4GB' is a GPU target; the file is 6.18 GB. Speeds, fit and picks.
Qwen 3.6-35B-A3B Is #1 Again: MiMo-V2.6-Flash Measured on an M5 Ultra
MiMo-V2.6-Flash's first eligible M5 Ultra run, 61.7 tok/s community-measured, not the ~79 estimated, makes Qwen 3.6-35B-A3B (24 GB) the index's #1 again.
MiniMax-M3.1-Flash on a Mac Studio: 256 GB, and MLX First
MiniMax's 428B M3.1-Flash runs on a 256 GB Mac Studio at 3-bit in MLX: community runs measured 34.5 tok/s on an M5 Ultra, 25.3 on an M3 Ultra.
Recursive Flint on a Mac: Speeds, RAM and Setup for the 40B-A2.5B Coding Model
Recursive Flint (40B, 2.5B active, Apache 2.0) on Apple Silicon: every measured tok/s by runtime, RAM from 24 to 48 GB, setup, and which Mac to buy.
Nemotron 3.5 Bolt on a Mac: The Metal Crash, What Works, Super VL
llama.cpp b11755 fixed Nemotron 3.5 Bolt's Metal crash; Ollama still can't run it on a Mac. What works, measured speeds, Bolt E4B and Super VL for 128 GB.
Muse Spark Open on a Mac: What Runs on One 512 GB Mac Studio
Meta's 816B Muse Spark Open fits one 512 GB Mac Studio at 2–3-bit: 21 tok/s (oQ3e) and 18 tok/s (Q2_K) published on M3 Ultra. 4-bit needs two Macs.
How Accurate Are LLMCheck's Mac Speed Estimates? 67 Runs Checked
LLMCheck's tok/s formula vs 67 new Mac runs: dense models within 18%, small MoEs right on average, large MoEs at 0.4×–2.1× the estimate.
Best Local LLM for a 24GB or 32GB Mac (October 2026)
According to the LLMCheck index, Qwen 3.6-35B-A3B is the best local LLM for 24GB and 32GB Macs. North Mini Code, KAT-Coder, LFM2.5 and MiniCPM-V by job.
Best Local LLM for a 16GB Mac (October 2026)
According to the LLMCheck index, Gemma 4 12B is the best local LLM for a 16GB Mac. Picks for coding, speed and ternary 27B models, with file sizes.
M5 Ultra vs M3 Ultra vs M5 Max for Local LLMs: The Measured Numbers
Published runs: the M5 Ultra decodes ~1.5× faster than an M3 Ultra and reads prompts 2–4× faster. When an M5 Max or M3 Ultra wins, and what fits 512 GB.
Best Local Vision Model for Mac (2026), by RAM Tier
MiniCPM-V 4.7 is the best local vision model for a 24–32 GB Mac. Picks from 8 GB to 512 GB, with memory needs, speeds and llama.cpp, MLX and Ollama setup.
State of Open-Source Local LLMs — October 2026: Small MoEs Take the Top
October 2026 report (5 Oct, updated 24 Oct): first Mac mini M6 runs, the MoE refit, Qwen 3.6-35B-A3B back at #1, Muse Spark Open, and Ollama goes MLX-first.
M5 Ultra and M6, Measured: How the LLMCheck Estimates Held Up (2026)
The first M5 Ultra and Mac mini M6 measurements set against the LLMCheck estimates: BGR, Apple, Tom's Hardware, tbreak and community runs, each labelled.
DeepSeek V4.1 Flash and MiMo-V2.6: Two MIT Frontier Models on a Mac (2026)
DeepSeek V4.1 Flash (552B, MIT) now runs on a 512 GB Mac Studio through oMLX; MiMo-V2.6-Flash (309B-A15B, MIT, omnimodal) runs from a 192 GB Mac Studio.
State of Open-Source Local LLMs — September 2026: Two Frontier MoEs Reach the Mac Studio
September 2026 open-source LLM report: Qwen3.8-27B takes #1, GLM-5.3-Flash (MIT, the Ox Alpha model) and Qwen3.8-Flash-Next land with Mac Studio-class quants, DeepSeek opens V4 Flash Vision, and the index recalibrates its MoE speed estimates.
GLM-5.3-Flash on a Mac: Which Studio Runs the Ox Alpha Model (2026)
GLM-5.3-Flash (320B-A18B, MIT) on a Mac: which quant fits 128 vs 256 GB, tok/s per chip, and running it in stock llama.cpp, oMLX or MLX.
Qwen3.8-Flash-Next on a Mac: 6B Active, 112 GB, and the License Clause to Read First (2026)
Qwen3.8-Flash-Next, the Qwen 4 architecture preview, needs a 128 GB Mac: 46 tok/s community-measured on an M5 Max. RAM, setup and the license clause.
Apple’s New Mac mini and Mac Studio: What M6 and M5 Ultra Mean for Local LLMs
Apple's M6 Mac mini and M5 Max and M5 Ultra Mac Studio for local LLMs: bandwidth, memory, prices and speeds from the LLMCheck index, with provenance.
M5 Ultra Mac Studio: 1.2 TB/s and 512 GB, Examined for Local LLMs
Apple's M5 Ultra for local LLMs: 1.2 TB/s, 50% above the M3 Ultra, and up to 512 GB. Review figures, index speeds and why 256 GB is the config to buy.
The $899 M6 Mac mini as a Local-AI Machine: What Actually Fits
Mac mini M6 for local LLMs: Qwen 3 30B-A3B at 62 tok/s (published), Qwen 3.6-35B-A3B at 64 (community), a dense 27B near 9. Which config to buy.
UI-Mate on Mac: RAM, Speed and Setup for Tencent’s Open Computer-Use Agent
UI-Mate-27B is Apache 2.0 and runs on Apple Silicon — but its 4-bit build is 17.5 GB and it holds five screenshots in context, so a 24 GB Mac is not enough. Real quant sizes, per-chip speed estimates, and which Mac actually works.
State of Open-Source Local LLMs — August 2026
August 2026 open-source LLM report: Meta returns with Apache-2.0 Muse Glimmer 30B, DeepSeek V4 Flash runs on 128GB Macs, Qwen open-weights its 2.4T Max, and the LLMCheck index completes a full provenance audit. Verified Mac #1: Qwen 3.6-27B.
GLM 5.2: The Open Model That Beat GPT-5 and Claude — on Its Own Harness
Zhipu AI's GLM 5.2 (June 2026, MIT, 753B-A40B MoE) reports 68.5% on SWE-Bench Pro — on its own scaffold. On Scale's standardized SEAL harness it scores ~62%, the top open-weights result. Harness-annotated benchmarks, plus the community quants that run it on 256–512 GB M3 Ultra Macs.
Apple Silicon Memory Bandwidth & LLM Speed (2026): M1–M5, Pro, Max, Ultra
Why memory bandwidth is the #1 factor in local LLM tokens-per-second on a Mac. Full GB/s table for every Apple Silicon chip (M1–M6, Pro/Max/Ultra) and how to estimate tok/s from bandwidth and model size.
Qwen 4 Release Date: What Alibaba Announced, and What to Run on a Mac
Qwen 4 is in training: announced 22 Sep 2026 with no weights, licence or date as of 24 Oct. What Alibaba confirmed, and which Qwen to run on a Mac now.
M4 vs M5 for Local LLMs: Is the New Apple Silicon Worth It?
M4 vs M5 Apple Silicon for local AI — what the M5's Neural Accelerators and higher bandwidth actually mean for tokens-per-second. Real benchmark deltas across M4/M5, Pro, and Max, and whether to upgrade.
State of Open-Source Local LLMs — July 2026 (Corrected Edition)
Corrected July 2026 open-source LLM report: GLM 5.2's harness-checked frontier lead, Thinking Machines' Inkling debut, Kimi K3's record 2.8T open weights, DeepSeek V4-Flash on 128 GB Macs, and the on-device ternary wave. Apple Silicon guidance with labeled provenance.
GLM 5.2 vs DeepSeek V4: The Real Open-Frontier Showdown
GLM 5.2 (753B, MIT) vs DeepSeek V4-Flash and V4 Pro — harness-annotated scores, license reality, and which open-frontier giant a Mac can actually run. Corrected: 'DeepSeek R3' never existed; the real DeepSeek line is R1 → V3.2 → V4.
GLM-4.5-Air on a Mac — and What Happened to "GLM 5.2 Air"
No 'GLM 5.2 Air' exists — the real Mac-runnable GLM with those specs is GLM-4.5-Air: 106B-A12B MoE, MIT licensed, ~30 tok/s (estimated) on a 64 GB Mac. Corrected setup guide, hardware requirements, and the newer GLM-4.7-Flash option.
GLM 5.2 vs Qwen3.8-Max vs Kimi K3: The Open Frontier Three-Way (August 2026)
GLM 5.2 (MIT), Qwen3.8-2.4T and Kimi K3 compared on licence, size and coding scores, and what a Mac can run instead, per the LLMCheck index.
Qwen 4 vs Qwen 4.1: What Exists, and Which Qwen to Run on a Mac
Qwen 4 was announced 22 Sep 2026 and is still in training, with no open weights; Qwen 4.1 does not exist. What shipped instead, and what to run on a Mac.
Phi-5 Doesn't Exist: Phi-4 Is Still Microsoft's Local Line (and What Beats It)
There is no Phi-5: Microsoft's local line tops out at Phi-4. Where Phi-4 14B still fits, and what the LLMCheck index picks for 24–32 GB Macs instead.
Llama 5 Still Isn't Out — Meta's Real 2026 Release Is Muse Glimmer 30B (August 2026)
Llama 5 has not been released. Meta's open model for 24–32 GB Macs is Muse Glimmer 30B: Apache 2.0, image input, 27 tok/s vendor-reported on an M5 Max.
State of Open-Source Local LLMs — June 2026
Corrected June 2026 report, kept as published with notes: its Qwen 4, Llama 5 70B, Voyage Pro, Grok 4 Open and Phi-5 Medium were not real open-weight releases.
Qwen 4 Coder Isn’t Out Yet: Qwen 4 Status and What to Run
Alibaba announced Qwen 4 on 22 Sep 2026; no Qwen 4 Coder weights exist yet. On a Mac, run Qwen 3.6-35B-A3B, KAT-Coder-V2.5 or Qwen3-Coder-Next instead.
Best 70B-Class Local Models (That Actually Exist): Llama 3.3 vs Hermes 4 vs Apertus 1.5
The 70B-class local models that actually exist, Llama 3.3, Hermes 4 and Apertus 1.5, compared on licence, memory and Mac speed per the LLMCheck index.
State of Open-Source Local LLMs — May 2026
Corrected May 2026 report, kept as published with notes: its Qwen 4 Preview, Llama 5, Phi-5 Mini and DeepSeek R2 were not real open-weight releases.
The Open Frontier, August 2026: Qwen3.8-Max vs Kimi K3 vs Inkling
The open-model frontier in August 2026: Qwen3.8's 2.4T open weights, Moonshot's 2.8T Kimi K3, and Thinking Machines' Inkling compared — benchmarks, licenses, and why none of them runs on a Mac (plus the Mac-class siblings that do).
DeepSeek V4 Pro vs Kimi K2.6: Best Open-Source Coding LLM in 2026
DeepSeek V4 Pro vs Kimi K2.6 Thinking on SWE-Bench, agentic coding, context and licence, and the local coding models a Mac runs instead.
Best Small LLMs for an 8GB Mac (August 2026)
The best small LLMs for an 8GB Mac in August 2026: LFM2.5-2.6B (vendor-reported 220 tok/s), ternary Maple Preview 20B-A1B and Bonsai 27B, Nanbeige4.2-3B, plus incumbents Phi-4 Mini and Gemma 4 E2B/E4B. Note: 'Phi-5 Mini', previously reviewed here, never shipped.
There Is No DeepSeek R2 — The Real DeepSeek Path for Macs
There is no DeepSeek R2. The DeepSeek a Mac can run is V4-Flash (MIT, 284B-A13B): 2-bit on a 128 GB Mac, ~39 tok/s community-measured on an M5 Max.
Qwen 3.6 vs Gemma 4 on a Mac: Speed, Coding and RAM
Qwen 3.6-35B-A3B vs Gemma 4 26B-A4B, 31B and 12B on a Mac: 118 vs 107 tok/s on an M5 Max in one published setup, 64 vs 48 on an M6, benchmarks and RAM.
Qwen 3.6-35B-A3B on Mac: 126 tok/s on an M5 Max, 64 on a Mac mini M6
The LLMCheck index's best pick for 24–32 GB Macs: 118 tok/s published and 126 community-measured on an M5 Max, 64 on a Mac mini M6. RAM, setup, benchmarks.
GLM-5.1: The First Open Model to Beat Claude on SWE-Bench Pro
GLM-5.1 (Z.ai, MIT, 744B MoE) scored 58.4% on SWE-Bench Pro in Z.ai's April 2026 report (vendor-run), above Claude Opus 4.6's 57.3%. What it means.
M5 Pro vs M5 Max for Local LLMs: 307 vs 614 GB/s (2026)
M5 Pro vs M5 Max for local AI: the Max has twice the bandwidth (614 vs 307 GB/s) and 128 GB for big MoE models. Speeds from the LLMCheck index.
M4 Max vs M3 Max for Local LLM: Is the Upgrade Worth It?
M4 Max vs M3 Max for local AI — is upgrading worth $400–600 more? The LLMCheck index shows ~35% faster tok/s on identical models. Full spec comparison and verdict.
How to Run Google Gemma 4 on Mac: Complete Setup Guide & Benchmarks
How to run Google Gemma 4 on Mac. Four variants from 2B to 31B parameters with Apache 2.0 licensing, 256K context, day-one MLX support, and multimodal input. Step-by-step Ollama setup and benchmark results.
Gemma 4 vs Qwen 3.5: Which Is the Best Local LLM for Mac in 2026?
Gemma 4 vs Qwen 3.5 head-to-head comparison on Apple Silicon. Benchmark data for tok/s, capability scores, RAM usage, multimodal support, and function calling across all model sizes.
Gemma 4 E2B & E4B: Run Google's AI on iPhone, iPad & Mac Mini
Gemma 4 E2B (2.3B params) and E4B (4B params) bring multimodal AI with audio, vision, and function calling to iPhone, iPad, and Mac Mini. Benchmarks, RAM requirements, and deployment guide.
Gemma 4 Hardware Requirements: RAM, M5 Chips & Apple Silicon Performance Guide
Gemma 4 hardware requirements for every Apple Silicon Mac. RAM needs from 1.5 GB (E2B) to 20 GB (31B) at INT4, tok/s benchmarks across M1 through M5 Ultra, and which Mac to buy for each Gemma 4 variant.
RAG vs Fine-Tuning on Mac: When to Use Each for Local AI
RAG vs fine-tuning for local AI on Mac. Decision framework comparing setup time, RAM, quality, cost, and use cases. Includes hybrid approach with MLX and Ollama.
Building AI Agents That Run Entirely on Your Mac in 2026
Guide to building local AI agents on Mac using CrewAI, AutoGen, and LangGraph with Ollama. No cloud APIs, no API keys — fully private agent frameworks.
Best Local LLM for Coding on Mac (October 2026)
According to the LLMCheck index: Mellum2.1 at 16 GB, North Mini Code or Qwen 3.6-35B-A3B at 24–32 GB, Qwen3-Coder-Next at 64 GB, with harness notes.
How Much RAM for Local AI on a Mac? 16GB to 512GB (2026)
How much RAM a Mac needs for local AI, from 8 GB to 512 GB: the LLMCheck index's pick for each tier, its speed on a named Mac, and what each step adds.
Running AI Without Internet: Complete Offline LLM Guide for Mac
How to run AI completely offline on your Mac. Download models once, then use them anywhere — planes, secure facilities, or areas without internet. Step-by-step guide with Ollama and LM Studio.
Llama 4 Scout on Mac: Setup Guide, Benchmarks & Performance
How to run Meta's Llama 4 Scout (109B MoE, 17B active) on a Mac: about 19 tok/s estimated on an M5 Max, needs 64 GB, 10M-token context. Ollama setup.
DeepSeek R1 vs Claude: Local vs Cloud AI for Developers
DeepSeek R1 running locally on Mac vs Claude in the cloud. A developer-focused comparison of reasoning quality, coding ability, privacy, cost, and speed with benchmark data.
Apple Silicon Neural Engine Explained: How Your Mac Runs AI
How Apple Silicon's Neural Engine, Unified Memory, and Metal GPU work together to run local AI. A technical explainer covering M1 through M5 architecture with performance data from LLMCheck benchmarks.
MoE vs Dense LLMs: Why MoE Models Run Faster on a Mac
MoE vs dense LLMs on a Mac: an MoE model needs RAM for all its parameters but reads only the active ones per token, so it decodes several times faster.
Llama 4 Scout & Maverick: Can You Run Meta's New AI on Your Mac?
Llama 4 Scout runs at about 19 tok/s estimated on an M5 Max (needs 64 GB); Maverick is server-only. Meta's MoE design, multimodal features, Ollama setup.
DeepSeek V3.2 vs GPT-5: Open Source Catches Up to Frontier AI
DeepSeek V3.2 scores 96% on AIME vs GPT-5's 94.6%. We compare benchmarks, architecture, licensing, and what this means for the open-source AI ecosystem in 2026.
M5 Max for Local AI (2026): Specs, Speeds, Model Picks
M5 Max decodes ~12% faster than the M4 Max (614 vs 546 GB/s) and reads prompts up to 4x faster (Apple). Specs, MLX setup and model picks for each M5 chip.
Qwen3-Coder-Next: Alibaba's Coding AI That Runs on Your Mac
Qwen3-Coder-Next scores 70.6% SWE-Bench with 80B MoE (3B active). About 49 tok/s estimated on a 64 GB Mac mini M5 Pro and 84 on an M5 Max. Review and setup.
7 Best Free Apps to Run AI Locally on Mac (2026 Guide)
The 7 best free apps to run AI locally on Mac in 2026: LM Studio, Ollama, Jan, Open WebUI, MLX, GPT4All, and Enchanted. Pros, cons, and install steps for each.
The Ultimate Interface Showdown: LM Studio vs Ollama for Mac (2026)
LM Studio vs Ollama for Mac in 2026 — which local LLM app is right for you? We compare setup, RAM usage, UI, and developer features to help you choose the best tool for your workflow.
M5 Max vs M4 Max for Local LLMs (2026): Worth the Upgrade?
M5 Max vs M4 Max for local LLMs: 614 vs 546 GB/s means ~12% faster decode; Apple quotes up to 4x faster prompt processing. Speeds by model, labelled.
Qwen 3.5 is Here: The Best Local LLM for Mac Just Changed Everything
Qwen 3.5 is the new best local LLM for Mac. Discover which model runs on your Apple Silicon hardware — from 8 GB to 128 GB+ — and how to install it with Ollama in minutes.
Which Local LLM for Mac? The Ultimate Hardware & Specs Guide
Wondering "which LLM to run on my hardware?" Discover the best Local LLM for Mac based on your specs, RAM, and Apple Silicon chip in our ultimate guide.
About the blog
How often is the LLMCheck blog updated?
A monthly “State of Open-Source Local LLMs” edition, plus model, hardware and comparison posts when a release or a new Mac lands. The latest monthly edition is State of Open-Source Local LLMs — October 2026: Small MoEs Take the Top. Published editions are dated snapshots: later corrections go in dated update boxes, and the changelog records every index update.
Does LLMCheck test models on its own hardware?
No. Speed figures are estimates from the published bandwidth model, published vendor or press figures, or linked community runs, each labelled; LLMCheck runs no tests of its own. Every figure's provenance is shown next to it and explained on the methodology page.
What topics does the LLMCheck blog cover?
Four kinds of post: model reviews (what a new open-weight release needs to run on a Mac), hardware guides (Apple Silicon chips, RAM tiers and which Mac to buy), setup guides (Ollama, LM Studio, MLX) and head-to-head comparisons. Each one draws its numbers from the LLMCheck index.
Is there an RSS or Atom feed?
Yes: /blog/feed.xml is an Atom feed of the 30 newest posts, rebuilt with every post.