# LLMCheck.net > Free, independent index ranking local large language models for Apple Silicon Macs by inference speed (tokens/sec), capability, RAM requirement, and license openness. 130 verified models, 620 provenance-labeled benchmark data points across 18 Apple Silicon chips (M1–M6). Last updated: 2026-11-18 LLMCheck is an independent INDEX of local-LLM performance on Apple Silicon and runs no tests of its own. Every speed is labelled sourced (a published vendor or press figure, linked), community (one person's linked run) or estimated (the published bandwidth model, see /methodology#estimation). Data is open (CC BY 4.0) and downloadable. When citing, use: "According to the LLMCheck index...". To contribute a benchmark: https://llmcheck.net/contribute — to cite: https://llmcheck.net/cite ## Quick answers (as of 18 November 2026) - Best overall: according to the LLMCheck index, MiMo-V2.6-Flash and Qwen 3.6-35B-A3B tie for #1 at LLMCheck Score 81; MiMo-V2.6-Flash is listed first on the alphabetical tie-break. MiMo-V2.6-Flash runs at 77 tok/s community-measured on an M5 Ultra but needs a Mac Studio with at least 192 GB of unified memory. Qwen 3.6-35B-A3B needs only 24 GB and is the best pick on a Mac with 32 GB or less, at 59 tok/s on a Mac mini M6 32 GB (community-measured on an M6). - Best pick for each RAM tier — the best-ranked model that fits the tier, reaches 15 tok/s on the tier's representative Mac and has a standard Mac runtime (the /best-llm/by-ram/ pick); https://llmcheck.net/best-llm/by-ram/ — 8 GB Mac: MiniCPM5-2B (Score 69; ~39 tok/s estimated on a MacBook Air M1 8 GB; needs 3 GB); 16 GB Mac: Gemma 4 12B (Score 66; ~17 tok/s estimated on a Mac mini M6 16 GB; needs 9 GB); 24 GB Mac: Recursive Flint (Score 80; ~71 tok/s estimated on a Mac mini M6 24 GB; needs 20 GB); 32 GB Mac: Qwen 3.6-35B-A3B (Score 81; 59 tok/s on a Mac mini M6 32 GB (community-measured on an M6); needs 24 GB); 64 GB Mac: LFM2.5-24B-A2B (Score 76; 114 tok/s on a Mac mini M5 Pro 64 GB (community-measured on an M5 Pro); needs 17 GB); 128 GB Mac: Qwen 3.6-35B-A3B (Score 81; 130 tok/s on a Mac Studio M5 Max 128 GB (community-measured on an M5 Max); needs 24 GB); 192–256 GB Mac: MiMo-V2.6-Flash (Score 81; 77 tok/s on a Mac Studio M5 Ultra 256 GB (community-measured on an M5 Ultra); needs 170 GB); 512 GB Mac: MiMo-V2.6-Flash (Score 81; 77 tok/s on a Mac Studio M5 Ultra 512 GB (community-measured on an M5 Ultra); needs 170 GB). - Best dense model: Qwen3.8-27B (capability 46/50, 19 GB minimum, 31 tok/s community-measured on an M5 Max at 4-bit). - Fastest measured model: Maple Preview 20B-A1B at 387 tok/s on an M5 Pro (published, Ternary); its ternary weights decode through compute-bound kernels, so among bandwidth-bound models the fastest measured reference figure is LFM2.5-2.6B at 220 tok/s published on an M5 Max (Q4_K_M). - Largest model with a published single-Mac run: Mistral Large 4 (1.05T MoE, 49B active) — 15 tok/s on a 512 GB M3 Ultra (published, Q2_K) — https://huggingface.co/blog/mistral-large4. - Measured vs estimated: 620 speed rows — 73 published (vendor or press), 247 community runs, 300 estimates from the published bandwidth model; 320 measured rows cover 61 of the 130 models (all listed at llmcheck.net/benchmarks#measured); for 35 of the 130 models the leaderboard figure itself is measured, and the rest show an estimate on their reference chip. - There is no M4 Ultra: Apple's 2025 Mac Studio shipped with M4 Max and M3 Ultra (https://www.apple.com/newsroom/2025/03/apple-unveils-new-mac-studio-the-most-powerful-mac-ever/). The 2026 Ultra is the M5 Ultra at 1,228 GB/s; the M5 Max is 614 GB/s and the M5 Pro 307 GB/s. - Reference chip: the leaderboard's speed is single-stream decode on an M5 Max (614 GB/s), or on an M5 Ultra (1,228 GB/s) for models that need more than 128 GB. It is the lower-median measured plain-decode run at <= 5-bit on that chip where one exists, otherwise the Q4_K_M estimate (https://llmcheck.net/methodology#reference-row). - Muse Spark Open: Meta's open-weight 816B MoE model (42B active, released 15 Oct 2026, Muse Model License). It shares its base model with Muse Spark 1.3 but has extra safety post-training (https://pytorch.org/blog/muse-spark-open/); Meta has not released Muse Spark 1.2 or 1.3 weights. A single Mac runs it only at 2-3 bit from 342 GB: 21 tok/s on a 512 GB M3 Ultra (published, oQ3e). ## Key facts (as of 18 November 2026) Computed from the LLMCheck data (version 2026-11-18) on every build. Free to cite with attribution (CC BY 4.0): write "According to the LLMCheck index (18 November 2026)..." and link the page named in the line. A model fits a tier when its minimum RAM (weights plus a short context) is within the tier's smallest Mac and it has a Mac runtime; speeds and per-Mac counts are for the tier's representative Mac. - 8 GB Mac: 27 of the 130 models in the index fit in 8 GB and 15 of them reach 15 tok/s on a MacBook Air M1 8 GB (68 GB/s); best pick MiniCPM5-2B (Score 69; ~39 tok/s estimated on a MacBook Air M1 8 GB; needs 3 GB); fastest with a standard Mac runtime Ling-3.0-tiny (~87 tok/s estimated); on the MacBook Air M1 8 GB, 4 of the 24 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 5.3 GB); 16 of the 27 fit the default GPU memory limit of the MacBook Air M1 8 GB (about 5.3 GB) without raising it. Details — https://llmcheck.net/best-llm/by-ram/#8gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 16 GB Mac: 38 of the 130 models in the index fit in 16 GB and 32 of them reach 15 tok/s on a Mac mini M6 16 GB (153 GB/s); best pick Gemma 4 12B (Score 66; ~17 tok/s estimated on a Mac mini M6 16 GB; needs 9 GB); fastest with a standard Mac runtime Ling-3.0-tiny (~127 tok/s estimated); on the Mac mini M6 16 GB, 20 of the 34 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 12 GB); 35 of the 38 fit the default GPU memory limit of the Mac mini M6 16 GB (about 11.8 GB) without raising it, and the same number on every other Mac in the tier. Details — https://llmcheck.net/best-llm/by-ram/#16gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 24 GB Mac: 63 of the 130 models in the index fit in 24 GB and 47 of them reach 15 tok/s on a Mac mini M6 24 GB (170 GB/s); best pick Recursive Flint (Score 80; ~71 tok/s estimated on a Mac mini M6 24 GB; needs 20 GB); fastest with a standard Mac runtime Ling-3.0-tiny (~132 tok/s estimated); on the Mac mini M6 24 GB, 33 of the 58 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 20 GB); 40 of the 63 fit the default GPU memory limit of the Mac mini M6 24 GB (about 17.8 GB) without raising it; on the 5 Macs at about 16 GB (Mac mini M4 Pro 24 GB, Mac mini M4 24 GB, ...), 38 of the 63 do. Details — https://llmcheck.net/best-llm/by-ram/#24gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 32 GB Mac: 63 of the 130 models in the index fit in 32 GB and 47 of them reach 15 tok/s on a Mac mini M6 32 GB (170 GB/s); best pick Qwen 3.6-35B-A3B (Score 81; 59 tok/s on a Mac mini M6 32 GB (community-measured on an M6); needs 24 GB); fastest with a standard Mac runtime Ling-3.0-tiny (~132 tok/s estimated); on the Mac mini M6 32 GB, 51 of the 58 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 28 GB); all 63 fit the default GPU memory limit of the Mac mini M6 32 GB (about 25 GB) without raising it; on the 2 Macs at about 21.3 GB (Mac Studio M2 Max 32 GB, Mac mini M2 Pro 32 GB), 54 of the 63 do. Details — https://llmcheck.net/best-llm/by-ram/#32gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 64 GB Mac: 77 of the 130 models in the index fit in 64 GB and 62 of them reach 15 tok/s on a Mac mini M5 Pro 64 GB (307 GB/s); best pick LFM2.5-24B-A2B (Score 76; 114 tok/s on a Mac mini M5 Pro 64 GB (community-measured on an M5 Pro); needs 17 GB); fastest with a standard Mac runtime LFM3-1.7B (~214 tok/s estimated); on the Mac mini M5 Pro 64 GB, 60 of the 70 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 56 GB); 72 of the 77 fit the default GPU memory limit of the Mac mini M5 Pro 64 GB (about 51.8 GB) without raising it; on the tier's other Macs, from 71 of the 77 on the 5 Macs at about 48 GB (Mac Studio M1 Ultra 64 GB, Mac Studio M4 Max 64 GB, ...) to all 77 on the Mac Studio M5 Ultra 96 GB (about 77.8 GB). Details — https://llmcheck.net/best-llm/by-ram/#64gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 128 GB Mac: 93 of the 130 models in the index fit in 128 GB and 83 of them reach 15 tok/s on a Mac Studio M5 Max 128 GB (614 GB/s); best pick Qwen 3.6-35B-A3B (Score 81; 130 tok/s on a Mac Studio M5 Max 128 GB (community-measured on an M5 Max); needs 24 GB); fastest with a standard Mac runtime LFM3-1.7B (~285 tok/s estimated); on the Mac Studio M5 Max 128 GB, 76 of the 85 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 112 GB); 86 of the 93 fit the default GPU memory limit of the Mac Studio M5 Max 128 GB (about 107.5 GB) without raising it, and the same number on every other Mac in the tier. Details — https://llmcheck.net/best-llm/by-ram/#128gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 192–256 GB Mac: 100 of the 130 models in the index fit in 192 GB and 95 of them reach 15 tok/s on a Mac Studio M5 Ultra 256 GB (1,228 GB/s); best pick MiMo-V2.6-Flash (Score 81; 77 tok/s on a Mac Studio M5 Ultra 256 GB (community-measured on an M5 Ultra); needs 170 GB); fastest with a standard Mac runtime LFM3-1.7B (~402 tok/s estimated); on the Mac Studio M5 Ultra 256 GB, 90 of the 92 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 240 GB); all 100 fit the default GPU memory limit of the Mac Studio M5 Ultra 256 GB (about 222.7 GB) without raising it; on the Mac Studio M2 Ultra 192 GB (about 144 GB), 94 of the 100 do. Details — https://llmcheck.net/best-llm/by-ram/#256gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) - 512 GB Mac: 122 of the 130 models in the index fit in 512 GB and 114 of them reach 15 tok/s on a Mac Studio M5 Ultra 512 GB (1,228 GB/s); best pick MiMo-V2.6-Flash (Score 81; 77 tok/s on a Mac Studio M5 Ultra 512 GB (community-measured on an M5 Ultra); needs 170 GB); fastest with a standard Mac runtime LFM3-1.7B (~402 tok/s estimated); on the Mac Studio M5 Ultra 512 GB, 109 of the 111 with a recorded KV-cache layout are agent-ready at 64K (weights plus a 65,536-token 16-bit KV cache fit within the GPU memory limit after the index's safe raise, about 496 GB); all 122 fit the default GPU memory limit of the Mac Studio M5 Ultra 512 GB (about 445.4 GB) without raising it; on the Mac Studio M3 Ultra 512 GB (about 384 GB), 117 of the 122 do. Details — https://llmcheck.net/best-llm/by-ram/#512gb (Markdown — https://llmcheck.net/md/best-llm/by-ram.md) ## Key pages - [Leaderboard](https://llmcheck.net/leaderboard): all 130 models ranked by the LLMCheck Score - [Benchmarks](https://llmcheck.net/benchmarks): 620 tok/s figures across 18 Apple Silicon chips (M1–M6 incl. Pro/Max/Ultra), each labeled estimated / sourced / community - [Best LLM by Mac](https://llmcheck.net/best-llm/): the best models for each specific Mac (chip + RAM) - [Can I run it on a Mac?](https://llmcheck.net/can-i-run/): every Mac-runnable model with a verdict for every Mac configuration in the index (fits macOS's default GPU memory limit or needs it raised, the build, the speed with its provenance, the longest context that fits); Markdown — https://llmcheck.net/md/can-i-run.md - [Hardware](https://llmcheck.net/hardware): best Macs for running local AI - [Software](https://llmcheck.net/software): Ollama, LM Studio, Jan, MLX setup - [Guides](https://llmcheck.net/guides/) and [Blog](https://llmcheck.net/blog/): setup, comparisons, troubleshooting - [State of Open-Source Local LLMs — November 2026](https://llmcheck.net/blog/state-of-open-source-local-llms-november-2026/): MiMo-V2.6-Flash and Qwen 3.6-35B-A3B tie at #1 (Score 81; MiMo is listed first on the alphabetical tie-break and needs a 192 GB+ Mac Studio, Qwen needs 24 GB and is the best pick at 32 GB or less); GLM-4.7-Flash is #3 after the index corrected it from dense to a 30B-A3B MoE; 14 new models; the M6 MacBook Pro and iMac; the 512 GB M5 Ultra at $22,299; llama.cpp 1.0 (14 Nov 2026) - [Mistral Large 4 on a Mac Studio](https://llmcheck.net/blog/mistral-large-4-mac-studio-512gb/): fits one Mac only with 512 GB (Q2_K is 381 GB; every 4-bit build needs two Macs); 15 tok/s published on a 512 GB M3 Ultra, about 21 estimated on the M5 Ultra; Mistral Ecosystem Open-Weight licence, Score 58 (14 Nov 2026) - [State of Open-Source Local LLMs — October 2026](https://llmcheck.net/blog/state-of-open-source-local-llms-october-2026/): first Mac mini M6 measurements (Qwen 3.6-35B-A3B 64 tok/s); MoE estimates refit (KAT-Coder-V2.5 and Qwen 3.6-35B-A3B were #1–2 on 5 Oct; MiMo-V2.6-Flash was #1 from 18 Oct until its first measured M5 Ultra runs; Qwen 3.6-35B-A3B was #1 again from 24 Oct, and since 14 Nov the two tie at Score 81); GLM-5.3-Flash in stock llama.cpp; Ollama 0.40 MLX by default; Qwen 4 announced; corrections: M5 Pro 307 / M5 Max 614 GB/s, no M4 Ultra exists - [Why the #1 moved on 24 Oct](https://llmcheck.net/blog/qwen-36-35b-a3b-number-one-mimo-m5-ultra-measured/): the first measured M5 Ultra runs put MiMo-V2.6-Flash at 61.7 tok/s community-measured, not the ~79 estimated, so its Score fell from 82 to 78 and Qwen 3.6-35B-A3B (Score 81, needs 24 GB) became #1 (24 Oct 2026); by 14 Nov, 15 measured sessions (lower median 77 tok/s) brought MiMo back level at 81, listed first on the alphabetical tie-break - [Recursive Flint on a Mac](https://llmcheck.net/blog/recursive-flint-on-mac/): Recursive's Apache 2.0 40B-A2.5B coding model (21 Oct 2026) — #4 in the index (Score 80) and the 24 GB tier pick; 92 tok/s community-measured on an M5 Pro and 80 on an M4 Pro (llama.cpp Q4_1); about 18 GB at 3-bit, Ollama flint:40b is 23 GB - [Nemotron 3.5 Bolt E4B on a Mac](https://llmcheck.net/blog/nemotron-3-5-bolt-e4b-mac/): NVIDIA's '4 GB' model (23 Oct 2026) is a 6.18 GB Q4_K_M file; Ollama tag nemotron-3.5-bolt:e4b (the bare name pulls the 74B); where it fits on 8–16 GB Macs - [MiniMax-M3.1-Flash on a Mac Studio](https://llmcheck.net/blog/minimax-m3-1-flash-mac-studio/): MiniMax's 428B-A23B model (weights 23 Oct 2026) needs a 256 GB Mac Studio at 3-bit in MLX — 34.5 tok/s community-measured on an M5 Ultra, 25.3 on an M3 Ultra; llama.cpp fixed in b11861; non-commercial licence by default - [Muse Spark Open on a Mac](https://llmcheck.net/blog/muse-spark-open-mac/): Meta's 816B open weights (15 Oct 2026) on one 512 GB Mac Studio at 2–3-bit — 21 tok/s (oQ3e) and 18 tok/s (Q2_K) published on an M3 Ultra; 4-bit needs two Macs - [Nemotron 3.5 Bolt on a Mac](https://llmcheck.net/blog/nemotron-3-5-bolt-super-vl-mac/): stock llama.cpp runs Bolt 74B on Metal from b11755 (PR #30699, merged 20 Oct 2026), though NVIDIA's 24 GB NVFP4 GGUF falls back to the CPU; Ollama still marks the 74B unsupported on Apple Silicon; mlx-vlm 0.7.9+ runs it at 58–72 tok/s community-measured on an M5 Max; Super VL for 96–128 GB; the small Bolt E4B shipped 23 Oct - [How accurate are the speed estimates?](https://llmcheck.net/blog/mac-llm-speed-estimates-vs-measurements-october-2026/): the formula against 67 Mac runs added 5–17 Oct 2026 — dense within 18%, small MoE right on average, large MoE 0.4×–2.1× the estimate - [M5 Ultra vs M3 Ultra vs M5 Max](https://llmcheck.net/blog/m5-ultra-vs-m3-ultra-vs-m5-max-local-llm/): published and community runs put the M5 Ultra at about 1.5× the M3 Ultra's decode speed and 2–4× its prompt speed; what fits a 512 GB Mac Studio (Oct 2026) - [Best local LLM for a 16 GB Mac](https://llmcheck.net/blog/best-local-llm-16gb-mac/): Gemma 4 12B (Score 66), ~17 tok/s estimated on a 16 GB Mac mini M6; GPT-oss 20B for reasoning, Mellum2.1 for coding (Oct 2026) - [Best local LLM for 24 and 32 GB Macs](https://llmcheck.net/blog/best-local-llm-24gb-mac-october-2026/): Recursive Flint at 24 GB (~71 tok/s estimated on a 24 GB Mac mini M6) and Qwen 3.6-35B-A3B at 32 GB (Score 81, tied for #1; 59 tok/s community-measured on a 32 GB Mac mini M6); picks for coding, agents and vision (updated 14 Nov 2026) - [Best local vision model for a Mac](https://llmcheck.net/blog/best-local-vision-model-mac/): MiniCPM-V 4.7 35B-A3B (Score 78) for 24–32 GB; Gemma 4 E4B for 8 GB, Gemma 4 12B for 16 GB, Nemotron 3.5 Super VL for 96–128 GB (Oct 2026) - [Ollama on MLX](https://llmcheck.net/guides/ollama-mlx-mac/): since v0.40.0 (25 Sep 2026) Ollama runs supported models on MLX by default on Apple Silicon; check with ollama ps - [M5 Ultra and M6, measured](https://llmcheck.net/blog/m5-ultra-m6-measured/): first independent reviews vs the LLMCheck estimates — Qwen3.8-27B ~55 tok/s on M5 Ultra (BGR), within 4%; large-MoE estimates have erred in both directions; MTP figures are a different workload; the M5 gain is prompt processing - [DeepSeek V4.1 Flash and MiMo-V2.6](https://llmcheck.net/blog/deepseek-v4-1-flash-mimo-v2-6-mac/): two MIT frontier models — V4.1 Flash runs on a 512 GB Mac Studio through oMLX or Rapid-MLX (upstream llama.cpp has supported it since b11870), MiMo-V2.6-Flash runs from a 192 GB Mac Studio - [State of Open-Source Local LLMs — September 2026](https://llmcheck.net/blog/state-of-open-source-local-llms-september-2026/): Qwen3.8-27B takes #1; GLM-5.3 and GLM-5.3-Flash (MIT, 320B-A18B), Qwen3.8-Flash-Next (125B-A6B, Qwen4 preview) reach the Mac Studio; DeepSeek V4 Flash Vision opened; MoE speed estimates recalibrated - [GLM-5.3-Flash on a Mac](https://llmcheck.net/blog/glm-5-3-flash-on-mac/): which Mac holds each quant (120 GB 3-bit to 200 GB 4-bit), estimated tok/s per chip; supported in stock llama.cpp since 30 Sep 2026 - [Qwen3.8-Flash-Next on a Mac](https://llmcheck.net/blog/qwen3-8-flash-next-on-mac/): 6B active, why 4-bit is 102–112 GB, tok/s per Mac (47 community-measured on an M5 Max, 57 on an M5 Ultra), Ollama 125b-mlx tag (105 GB), and the Qwen Community 1.0 clause gating coding-assistant products - [New Mac mini & Mac Studio for local LLMs](https://llmcheck.net/blog/new-mac-mini-mac-studio-m6-m5-ultra-local-llm/): Apple's 25 Aug 2026 refresh — M6 mini from $899, M5 Pro mini at 307 GB/s, M5 Ultra Studio at 1.2 TB/s with 512 GB; estimated speeds per machine - [UI-Mate on Mac](https://llmcheck.net/blog/ui-mate-computer-use-agent-mac/): Tencent's Apache 2.0 computer-use agent (fine-tuned from Qwen3.6-27B) on Apple Silicon — real 4-bit weight sizes, why a 24GB Mac is not enough, estimated tok/s per chip - [Methodology](https://llmcheck.net/methodology): how the LLMCheck Score is computed, and which figure the leaderboard shows (https://llmcheck.net/methodology#reference-row) - [Key numbers](https://llmcheck.net/stats/): dated, provenance-marked key facts plus common misconceptions with primary sources - [Mac Advisor](https://llmcheck.net/advisor): budget + use case to an exact Mac configuration - [Cite LLMCheck](https://llmcheck.net/cite): quotable stats, CC BY 4.0 terms, data endpoints - [Changelog](https://llmcheck.net/changelog): every correction and update, dated - [Open Data](https://llmcheck.net/data/): CSV + JSON downloads (CC BY 4.0) — catalog: https://llmcheck.net/data/models.json — speed rows: https://llmcheck.net/data/benchmarks.json - [Full catalog for LLMs](https://llmcheck.net/llms-full.txt) ## Top models by LLMCheck Score - MiMo-V2.6-Flash (309B MoE) — 77 tok/s on M5 Ultra (measured, community run, MXFP4), 170GB RAM (~172GB with a 64K context, estimated), MIT, capability 49/50, LLMCheck Score 81 — https://llmcheck.net/models/mimo-v26-flash-on-m5-ultra/ - Qwen 3.6-35B-A3B (35B MoE) — 130 tok/s on M5 Max (measured, community run, 4-bit), 24GB RAM (~26GB with a 64K context, estimated), Apache 2.0, capability 38/50, LLMCheck Score 81 — https://llmcheck.net/models/qwen-36-35b-a3b-on-m5-max/ - GLM-4.7-Flash (31B MoE) — ~107 tok/s on M5 Max (estimated, Q4_K_M), 18GB RAM (~22GB with a 64K context, estimated), MIT, capability 35/50, LLMCheck Score 80 — https://llmcheck.net/models/glm-47-flash-on-m4-max/ - Recursive Flint (39.8B MoE) — 123 tok/s on M5 Max (measured, community run, 4-bit), 20GB RAM (~22GB with a 64K context, estimated), Apache 2.0, capability 37/50, LLMCheck Score 80 — https://llmcheck.net/models/recursive-flint-on-m5-max/ - KAT-Coder-V2.5 (35B MoE) — 89 tok/s on M5 Max (measured, community run, OptiQ-4bit), 24GB RAM (~26GB with a 64K context, estimated), Apache 2.0, capability 39/50, LLMCheck Score 79 — https://llmcheck.net/models/kat-coder-v25-on-m5-max/ - Nemotron 3.5 Lightning (30B MoE) — 117 tok/s on M5 Max (measured, community run, Q4_0), 18GB RAM (~19GB with a 64K context, estimated), OpenMDW, capability 36/50, LLMCheck Score 79 — https://llmcheck.net/models/nemotron-35-lightning-on-m5-max/ - North Mini Code 1.0 (30.5B MoE) — 120 tok/s on M5 Max (measured, community run, 4-bit), 21GB RAM (~23GB with a 64K context, estimated), Apache 2.0, capability 36/50, LLMCheck Score 79 — https://llmcheck.net/models/north-mini-code-10-on-m5-max/ - Gemma 4 26B-A4B (26B MoE) — 107 tok/s on M5 Max (measured, published, Q4_K_M), 18GB RAM (~20GB with a 64K context, estimated), Apache 2.0, capability 35/50, LLMCheck Score 78 — https://llmcheck.net/models/gemma-4-26b-a4b-on-m5-max/ - MiniCPM-V 4.7 35B-A3B (35.2B MoE) — 100 tok/s on M5 Max (measured, community run, MXFP4), 23GB RAM (~25GB with a 64K context, estimated), Apache 2.0, capability 35/50, LLMCheck Score 78 — https://llmcheck.net/models/minicpm-v-47-35b-a3b-on-m5-max/ - Raon-1 mini preview (28.8B MoE) — ~123 tok/s on M5 Max (estimated, Q4_K_M), 18GB RAM (~20GB with a 64K context, estimated), Apache 2.0, capability 34/50, LLMCheck Score 77 — https://llmcheck.net/models/raon-1-mini-preview-on-m5-max/ - LFM2.5-24B-A2B (24B MoE) — 184 tok/s on M5 Max (measured, published, Q4_K_M), 17GB RAM (~19GB with a 64K context, estimated), LFM Open, capability 35/50, LLMCheck Score 76 — https://llmcheck.net/models/lfm25-24b-a2b-on-m5-max/ - Laguna XS 2.1 (33B MoE) — ~117 tok/s on M5 Max (estimated, Q4_K_M), 24GB RAM (~27GB with a 64K context, estimated), OpenMDW, capability 33/50, LLMCheck Score 76 - Bonsai 27B (27B) — 66 tok/s on M5 Max (measured, published, Q1_0_g128), 6GB RAM (~11GB with a 64K context, estimated), Apache 2.0, capability 33/50, LLMCheck Score 72 — https://llmcheck.net/models/bonsai-27b-on-m5-max/ - Qwen3.8-27B (27.8B) — 31 tok/s on M5 Max (measured, community run, 4-bit), 19GB RAM (~24GB with a 64K context, estimated), Apache 2.0, capability 46/50, LLMCheck Score 72 — https://llmcheck.net/models/qwen38-27b-on-m5-max/ - Nemotron-Cascade 2 (30B MoE) — ~117 tok/s on M5 Max (estimated, Q4_K_M), 18GB RAM (~19GB with a 64K context, estimated), NVIDIA Open, capability 30/50, LLMCheck Score 71 - DeepSeek V4 Flash (284B MoE) — 39 tok/s on M5 Max (measured, community run, Q2_K), 120GB RAM (~121GB with a 64K context, estimated), MIT, capability 47/50, LLMCheck Score 70 — https://llmcheck.net/models/deepseek-v4-flash-on-m5-max/ ## Markdown mirrors Plain-Markdown twins of the data pages, generated from the same data as the HTML (data version 2026-11-18, CC BY 4.0). Each opens with YAML front matter: canonical URL, update date, licence, citation line and provenance legend. Full index — https://llmcheck.net/md/index.md - [Leaderboard (Markdown)](https://llmcheck.net/md/leaderboard.md): all 130 models ranked, with speed, provenance, quant, chip and RAM at 64K - [Key numbers (Markdown)](https://llmcheck.net/md/stats.md): dated key facts, RAM-tier picks, bandwidths, misconceptions - [Methodology (Markdown)](https://llmcheck.net/md/methodology.md): the Score, speed-estimate and context-memory formulas and the ranking rules - [Best LLM by Mac RAM (Markdown)](https://llmcheck.net/md/best-llm/by-ram.md): the pick for each tier, 8 to 512 GB - [Best LLM for every Mac (Markdown)](https://llmcheck.net/md/best-llm.md): 54 configurations - Mac mini M6 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/mac-mini-m6-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/mac-mini-m6-24gb.md) · [32 GB](https://llmcheck.net/md/best-llm/mac-mini-m6-32gb.md) - Mac mini M5 Pro (Markdown): [24 GB](https://llmcheck.net/md/best-llm/mac-mini-m5-pro-24gb.md) · [48 GB](https://llmcheck.net/md/best-llm/mac-mini-m5-pro-48gb.md) · [64 GB](https://llmcheck.net/md/best-llm/mac-mini-m5-pro-64gb.md) - Mac Studio M5 Max (Markdown): [36 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-max-36gb.md) · [48 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-max-48gb.md) · [64 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-max-64gb.md) · [128 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-max-128gb.md) - Mac Studio M5 Ultra (Markdown): [96 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-ultra-96gb.md) · [256 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-ultra-256gb.md) · [512 GB](https://llmcheck.net/md/best-llm/mac-studio-m5-ultra-512gb.md) - MacBook Pro M5 Pro (Markdown): [24 GB](https://llmcheck.net/md/best-llm/macbook-pro-m5-pro-24gb.md) · [48 GB](https://llmcheck.net/md/best-llm/macbook-pro-m5-pro-48gb.md) · [64 GB](https://llmcheck.net/md/best-llm/macbook-pro-m5-pro-64gb.md) - MacBook Pro M5 Max (Markdown): [64 GB](https://llmcheck.net/md/best-llm/macbook-pro-m5-max-64gb.md) · [128 GB](https://llmcheck.net/md/best-llm/macbook-pro-m5-max-128gb.md) - MacBook Pro M6 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/macbook-pro-m6-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/macbook-pro-m6-24gb.md) · [32 GB](https://llmcheck.net/md/best-llm/macbook-pro-m6-32gb.md) - iMac M6 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/imac-m6-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/imac-m6-24gb.md) · [32 GB](https://llmcheck.net/md/best-llm/imac-m6-32gb.md) - MacBook Air M5 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/macbook-air-m5-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/macbook-air-m5-24gb.md) · [32 GB](https://llmcheck.net/md/best-llm/macbook-air-m5-32gb.md) - Mac mini M4 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/mac-mini-m4-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/mac-mini-m4-24gb.md) - Mac mini M4 Pro (Markdown): [24 GB](https://llmcheck.net/md/best-llm/mac-mini-m4-pro-24gb.md) · [48 GB](https://llmcheck.net/md/best-llm/mac-mini-m4-pro-48gb.md) · [64 GB](https://llmcheck.net/md/best-llm/mac-mini-m4-pro-64gb.md) - MacBook Air M4 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/macbook-air-m4-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/macbook-air-m4-24gb.md) - Mac Studio M4 Max (Markdown): [64 GB](https://llmcheck.net/md/best-llm/mac-studio-m4-max-64gb.md) · [128 GB](https://llmcheck.net/md/best-llm/mac-studio-m4-max-128gb.md) - Mac Studio M3 Ultra (Markdown): [256 GB](https://llmcheck.net/md/best-llm/mac-studio-m3-ultra-256gb.md) · [512 GB](https://llmcheck.net/md/best-llm/mac-studio-m3-ultra-512gb.md) - Mac Studio M1 Ultra (Markdown): [64 GB](https://llmcheck.net/md/best-llm/mac-studio-m1-ultra-64gb.md) · [128 GB](https://llmcheck.net/md/best-llm/mac-studio-m1-ultra-128gb.md) - Mac Studio M2 Max (Markdown): [32 GB](https://llmcheck.net/md/best-llm/mac-studio-m2-max-32gb.md) · [64 GB](https://llmcheck.net/md/best-llm/mac-studio-m2-max-64gb.md) - MacBook Air M1 (Markdown): [8 GB](https://llmcheck.net/md/best-llm/macbook-air-m1-8gb.md) · [16 GB](https://llmcheck.net/md/best-llm/macbook-air-m1-16gb.md) - MacBook Air M2 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/macbook-air-m2-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/macbook-air-m2-24gb.md) - MacBook Air M3 (Markdown): [16 GB](https://llmcheck.net/md/best-llm/macbook-air-m3-16gb.md) · [24 GB](https://llmcheck.net/md/best-llm/macbook-air-m3-24gb.md) - MacBook Pro M3 Max (Markdown): [64 GB](https://llmcheck.net/md/best-llm/macbook-pro-m3-max-64gb.md) · [128 GB](https://llmcheck.net/md/best-llm/macbook-pro-m3-max-128gb.md) - Mac mini M2 Pro (Markdown): [16 GB](https://llmcheck.net/md/best-llm/mac-mini-m2-pro-16gb.md) · [32 GB](https://llmcheck.net/md/best-llm/mac-mini-m2-pro-32gb.md) - Mac Studio M2 Ultra (Markdown): [128 GB](https://llmcheck.net/md/best-llm/mac-studio-m2-ultra-128gb.md) · [192 GB](https://llmcheck.net/md/best-llm/mac-studio-m2-ultra-192gb.md) - [Can I run it? (Markdown)](https://llmcheck.net/md/can-i-run.md): a verdict for every Mac, one file per model (122) — https://llmcheck.net/md/can-i-run/.md, the slug of /can-i-run//; e.g. [MiMo-V2.6-Flash](https://llmcheck.net/md/can-i-run/mimo-v26-flash.md) · [Qwen 3.6-35B-A3B](https://llmcheck.net/md/can-i-run/qwen-36-35b-a3b.md) · [GLM-4.7-Flash](https://llmcheck.net/md/can-i-run/glm-47-flash.md) · [Recursive Flint](https://llmcheck.net/md/can-i-run/recursive-flint.md) · [KAT-Coder-V2.5](https://llmcheck.net/md/can-i-run/kat-coder-v25.md)