# LLMCheck.net — full catalog

> 85 verified local LLMs indexed for Apple Silicon Macs. Fields: name, size, min RAM (GB), speed (tok/s on M5 Max; estimated unless noted sourced/community in the dataset), license, LLMCheck capability score (/50). Source: https://llmcheck.net/data/ (CC BY 4.0). Catalog audited August 2026 — every entry verified against official sources.

## All models
- GLM 5.2 (753B MoE) — server-class, 256GB RAM, MIT, capability 50/50
- DeepSeek V4 Pro (1.6T MoE) — server-class, 850GB RAM, MIT, capability 50/50
- Qwen3.8-2.4T-A95B (2.4T MoE) — server-class, 1310GB RAM, Qwen Custom, capability 50/50
- Kimi K3 (2.8T MoE) — server-class, 1560GB RAM, Moonshot Custom, capability 50/50
- GLM-5.1 (754B MoE) — server-class, 390GB RAM, MIT, capability 48/50
- Kimi K2.6 (1T MoE) — server-class, 600GB RAM, Modified MIT, capability 48/50
- DeepSeek V4 Flash (284B MoE) — ~39 tok/s, 120GB RAM, MIT, capability 47/50
- Inkling (975B MoE) — server-class, 550GB RAM, Apache 2.0, capability 47/50
- Kimi K2.5 (1T MoE) — server-class, 600GB RAM, Modified MIT, capability 47/50
- Qwen3.8-27B (27.8B) — ~29 tok/s (estimated), 19GB RAM, Apache 2.0, capability 46/50
- GLM-5.3 (753B MoE) — ~29 tok/s (estimated), 256GB RAM, GLM-5.3 License, capability 50/50
- GLM-5.3-Flash (320B MoE) — ~27 tok/s (estimated), 128GB RAM, MIT, capability 48/50
- Qwen3.8-Flash-Next (125B MoE) — ~49 tok/s (estimated), 120GB RAM, Qwen Custom, capability 47/50
- DeepSeek V4 Flash Vision-Exp (305B MoE) — ~39 tok/s (estimated), 120GB RAM, MIT, capability 47/50
- Inkling-Small (276B MoE) — ~22 tok/s, 96GB RAM, Apache 2.0, capability 46/50
- Qwen3-235B-A22B (235B MoE) — ~15 tok/s, 128GB RAM, Apache 2.0, capability 46/50
- DeepSeek V3.2-Speciale (685B MoE) — server-class, 360GB RAM, MIT, capability 45/50
- Qwen 3.6-27B (27B) — ~30 tok/s, 18GB RAM, Apache 2.0, capability 44/50
- Laguna S 2.1 (118B MoE) — ~26 tok/s, 96GB RAM, OpenMDW, capability 43/50
- Muse Glimmer 30B (30B) — ~27 tok/s, 18GB RAM, Apache 2.0, capability 42/50
- Solar Open 2 250B (250B MoE) — server-class, 160GB RAM, Solar License, capability 42/50
- DeepSeek V3.2 (685B MoE) — server-class, 360GB RAM, MIT, capability 42/50
- Hunyuan Hy3 (295B MoE) — server-class, 160GB RAM, Apache 2.0, capability 41/50
- Gemma 4 31B (31B) — ~26 tok/s, 24GB RAM, Apache 2.0, capability 40/50
- MiniMax M2.5 (230B MoE) — server-class, 130GB RAM, Modified MIT, capability 40/50
- KAT-Coder-V2.5 (35B MoE) — ~52 tok/s, 24GB RAM, Apache 2.0, capability 39/50
- Qwen 3.6-35B-A3B (35B MoE) — ~52 tok/s, 24GB RAM, Apache 2.0, capability 38/50
- Ling-3.0 Flash (124B MoE) — server-class, 96GB RAM, MIT, capability 38/50
- Qwen 3.5 397B-A17B (397B MoE) — server-class, 210GB RAM, Apache 2.0, capability 38/50
- DeepSeek R1 (671B MoE) — server-class, 350GB RAM, MIT, capability 37/50
- DeepSeek V3 (685B MoE) — server-class, 360GB RAM, MIT, capability 37/50
- Nemotron 3.5 Lightning (30B MoE) — ~60 tok/s, 18GB RAM, OpenMDW, capability 36/50
- Llama 4 Maverick (400B MoE) — server-class, 210GB RAM, Meta Custom, capability 36/50
- Mistral Large 3 (675B MoE) — server-class, 355GB RAM, Apache 2.0, capability 36/50
- GLM-4.7-Flash (31B) — ~26 tok/s, 18GB RAM, MIT, capability 35/50
- Gemma 4 26B-A4B (26B MoE) — ~48 tok/s, 18GB RAM, Apache 2.0, capability 35/50
- Qwen3-Coder-Next (80B MoE) — ~35 tok/s, 48GB RAM, Apache 2.0, capability 35/50
- Llama 3.1 405B (405B) — server-class, 240GB RAM, Meta Custom, capability 35/50
- Mistral Small 4 (119B MoE) — ~38 tok/s, 64GB RAM, Apache 2.0, capability 34/50
- GLM-4.7 (355B) — server-class, 180GB RAM, MIT, capability 34/50
- Bonsai 27B (27B) — ~30 tok/s, 6GB RAM, Apache 2.0, capability 33/50
- Laguna XS 2.1 (33B MoE) — ~55 tok/s, 24GB RAM, OpenMDW, capability 33/50
- GLM-4.5-Air (106B MoE) — ~30 tok/s, 56GB RAM, MIT, capability 33/50
- Step-3.5-Flash (196B MoE) — ~18 tok/s, 120GB RAM, Apache 2.0, capability 33/50
- GPT-oss 120B (117B) — ~7 tok/s, 62GB RAM, Apache 2.0, capability 32/50
- MiMo-V2-Flash (309B MoE) — server-class, 165GB RAM, MIT, capability 32/50
- Nemotron-Cascade 2 (30B MoE) — ~58 tok/s, 18GB RAM, NVIDIA Open, capability 30/50
- DeepSeek R1 70B (70B) — ~12 tok/s, 44GB RAM, MIT, capability 30/50
- Llama 4 Scout (109B MoE) — ~32 tok/s, 58GB RAM, Meta Custom, capability 30/50
- Apertus 1.5 70B (70B) — ~12 tok/s, 44GB RAM, Apache 2.0, capability 29/50
- Hermes 4 70B (70B) — ~12 tok/s, 38GB RAM, Meta Custom, capability 28/50
- Qwen 2.5 72B (72B) — ~12 tok/s, 46GB RAM, Apache 2.0, capability 28/50
- Mixtral 8x22B (141B MoE) — ~15 tok/s, 88GB RAM, Apache 2.0, capability 28/50
- Qwen 3.5 35B-A3B (35B MoE) — ~45 tok/s, 24GB RAM, Apache 2.0, capability 27/50
- Llama 3.3 70B (70B) — ~12 tok/s, 44GB RAM, Meta Custom, capability 27/50
- QwQ 32B (32B) — ~25 tok/s, 22GB RAM, Apache 2.0, capability 26/50
- Mistral Small 3.2 24B (24B) — ~33 tok/s, 14GB RAM, Apache 2.0, capability 25/50
- Qwen 3 32B (32B) — ~25 tok/s, 22GB RAM, Apache 2.0, capability 25/50
- Devstral Small 24B (24B) — ~33 tok/s, 14GB RAM, Apache 2.0, capability 24/50
- DeepSeek R1 32B (32B) — ~25 tok/s, 22GB RAM, MIT, capability 24/50
- Qwen 3 30B-A3B (30B MoE) — ~58 tok/s, 18GB RAM, Apache 2.0, capability 22/50
- Gemma 3 27B (27B) — ~30 tok/s, 18GB RAM, Gemma, capability 22/50
- Qwen 3.5 27B (27B) — ~30 tok/s, 18GB RAM, Apache 2.0, capability 21/50
- Nanbeige4.2-3B (3B) — ~199 tok/s, 2GB RAM, Apache 2.0, capability 20/50
- Maple Preview 20B-A1B (20B MoE) — ~281 tok/s, 6GB RAM, MIT, capability 20/50
- Qwen 3 14B (14B) — ~55 tok/s, 10GB RAM, Apache 2.0, capability 20/50
- Phi-4 14B (14B) — ~55 tok/s, 10GB RAM, MIT, capability 19/50
- Qwen 3.5 9B (9B) — ~82 tok/s, 7GB RAM, Apache 2.0, capability 18/50
- Qwen 2.5 14B (14B) — ~55 tok/s, 10GB RAM, Apache 2.0, capability 18/50
- Ministral 3 14B (14B) — ~55 tok/s, 10GB RAM, Apache 2.0, capability 18/50
- Gemma 3 12B (12B) — ~64 tok/s, 9GB RAM, Gemma, capability 17/50
- Gemma 4 E4B (4B) — ~161 tok/s, 3GB RAM, Apache 2.0, capability 16/50
- Apertus 1.5 8B (8B) — ~91 tok/s, 5GB RAM, Apache 2.0, capability 16/50
- DeepSeek R1 8B (8B) — ~91 tok/s, 6GB RAM, MIT, capability 16/50
- Qwen 3 8B (8B) — ~91 tok/s, 6GB RAM, Apache 2.0, capability 15/50
- LFM2.5-2.6B (2.6B) — ~220 tok/s, 2GB RAM, LFM Open, capability 14/50
- Phi-4 Mini (3.8B) — ~168 tok/s, 3GB RAM, MIT, capability 14/50
- Ministral 8B (8B) — ~91 tok/s, 6GB RAM, Apache 2.0, capability 14/50
- Gemma 4 E2B (2B) — ~261 tok/s, 2GB RAM, Apache 2.0, capability 13/50
- Mistral 7B (7B) — ~102 tok/s, 5GB RAM, Apache 2.0, capability 13/50
- Qwen 3.5 4B (4B) — ~161 tok/s, 3GB RAM, Apache 2.0, capability 12/50
- Llama 3.1 8B (8B) — ~91 tok/s, 5GB RAM, Meta Custom, capability 12/50
- Qwen 3 4B (4B) — ~161 tok/s, 3GB RAM, Apache 2.0, capability 11/50
- SmolLM3 3B (3B) — ~199 tok/s, 2GB RAM, Apache 2.0, capability 10/50
- Gemma 3 4B (4B) — ~161 tok/s, 3GB RAM, Gemma, capability 10/50
- [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