{"id":1534,"date":"2026-09-18T06:21:58","date_gmt":"2026-09-17T21:21:58","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/home-en\/"},"modified":"2026-10-27T09:08:33","modified_gmt":"2026-10-27T00:08:33","slug":"home-en","status":"publish","type":"page","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/","title":{"rendered":"Local LLM News, Quick References and Our Own Measurements"},"content":{"rendered":"

Local Model Watch is a reference site for open-weight generative models (text, image, video and audio) you can run on your own hardware, in English and Japanese: articles on new releases, hub pages by VRAM, model family, publisher, format and inference engine, and measurements we take ourselves by running models on our own server. Memory requirements in every article are computed by this site from the actual distributed file sizes \u2014 see our editorial policy<\/a>.<\/p>\n

Start here<\/h2>\n
Find models that fit your GPU<\/strong>
Local models by VRAM tier \u2014 52 models listed<\/span><\/a>See our own measurements<\/strong>
Japanese token efficiency, GGUF internals, CPU runs and generated media for 47 models<\/span><\/a>Browse by model family<\/strong>
17 families with articles, hardware tables and tracked variants<\/span><\/a>See what’s trending<\/strong>
Likes and downloads of 70 models we covered, updated weekly<\/span><\/a>Browse by publisher<\/strong>
49 companies and labs: their Hugging Face models and our coverage<\/span><\/a>Browse by model format<\/strong>
8 formats such as GGUF, MLX and FP8: what they are and which models ship in them<\/span><\/a>Follow inference engines and tools<\/strong>
Release history of 25 projects (llama.cpp, Ollama, vLLM, …)<\/span><\/a>Browse by task<\/strong>
Image, audio, video and text models side by side \u2014 65 models<\/span><\/a>Read the glossaries<\/strong>
45 benchmarks explained, plus quantization and model formats<\/span><\/a><\/div>\n

Everything that fits your GPU: 8GB<\/a> \/ 12GB<\/a> \/ 16GB<\/a> \/ 24GB<\/a><\/p>\n

Quantization and model-format glossary<\/a> is also available.<\/p>\n

Our latest measurements<\/h2>\n

Models we ran ourselves on our server’s CPU (Neoverse-N1, no GPU) with the official llama.cpp builds. These are CPU figures, not a guide to GPU speed. Each article shows the answers verbatim.<\/p>\n

\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n\n
Model<\/th>\nFile (quant)<\/th>\nGeneration<\/th>\nIn Japanese<\/th>\nPeak memory<\/th>\nConditions<\/th>\nMeasured<\/th>\n<\/tr>\n<\/thead>\n
clef-flash Vision-Language Model: Our Test Answers, 4GB+ VRAM<\/a><\/td>\nCloudflare_clef-flash-Q4_K_M.gguf<\/code>(Q4_K_M<\/code>)<\/td>\n3.9 tok\/s<\/td>\n7.1 chars\/s<\/td>\n11.8GB<\/td>\nllama.cpp b11347, 3 threads<\/td>\n2026-10-03<\/td>\n<\/tr>\n
MiMo-V2.6-Distill-Qwen-9B-GGUF: Our Test Answers, 12GB+ VRAM<\/a><\/td>\nMiMo-V2.6-Distill-Qwen-9B-Q8_0.gguf<\/code>(Q8_0<\/code>)<\/td>\n3.7 tok\/s<\/td>\n6.8 chars\/s<\/td>\n14.8GB<\/td>\nllama.cpp b11223, 3 threads<\/td>\n2026-09-30<\/td>\n<\/tr>\n
Nex-N2.5-mini Text Generation Model: Our Test Answers, 16GB+ VRAM<\/a><\/td>\nnex-agi_Nex-N2.5-mini-Q2_K.gguf<\/code>(Q2_K<\/code>)<\/td>\n7.3 tok\/s<\/td>\n13.4 chars\/s<\/td>\n14.9GB<\/td>\nllama.cpp b11223, 3 threads<\/td>\n2026-09-29<\/td>\n<\/tr>\n
Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-GGUF: Our Test Answers<\/a><\/td>\nGryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-Q3_K_S.gguf<\/code>(Q3_K_S<\/code>)<\/td>\n6.2 tok\/s<\/td>\n11.0 chars\/s<\/td>\n14.9GB<\/td>\nllama.cpp b11223, 3 threads<\/td>\n2026-09-29<\/td>\n<\/tr>\n
vectionlabs_Salience-27B-R6-GGUF: Our Test Answers, 12GB+ VRAM<\/a><\/td>\nvectionlabs_Salience-27B-R6-Q3_K_M.gguf<\/code>(Q3_K_M<\/code>)<\/td>\n0.9 tok\/s<\/td>\n1.6 chars\/s<\/td>\n14.7GB<\/td>\nllama.cpp b11223, 3 threads<\/td>\n2026-09-29<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Models our CPU cannot run were run on a rented cloud GPU. Speeds are for that one GPU and engine, not for llama.cpp or other GPUs.<\/p>\n

\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n
Model<\/th>\nEngine<\/th>\nGPU<\/th>\nVRAM after loading<\/th>\nPeak VRAM<\/th>\nGeneration<\/th>\nMeasured<\/th>\n<\/tr>\n<\/thead>\n
ELYZA Releases ELYZA-Thinking-1.0 32B\/33B Reasoning Models<\/a><\/td>\ntransformers<\/td>\nNVIDIA A100-SXM4-80GB<\/td>\n59.9GB<\/td>\n60.1GB<\/td>\n13.7 tok\/s<\/td>\n2026-10-11<\/td>\n<\/tr>\n
K2-Horizon-32B-NVFP4: Our Test Answers, 32GB+ VRAM<\/a><\/td>\ntransformers<\/td>\nNVIDIA A100-SXM4-80GB<\/td>\n21.7GB<\/td>\n71.3GB<\/td>\n2.2 tok\/s<\/td>\n2026-10-02<\/td>\n<\/tr>\n
OrcaSAQ-2-27B Text Generation Model: Our Test Answers, 16GB+ VRAM<\/a><\/td>\nExLlamaV3 1.5.3<\/td>\nNVIDIA L4<\/td>\n13.7GB<\/td>\n14.0GB<\/td>\n17.2 tok\/s<\/td>\n2026-10-01<\/td>\n<\/tr>\n
Xing4.0-29B-A4B Text Generation Model: Our Test Answers, 24GB+ VRAM<\/a><\/td>\ntransformers<\/td>\nNVIDIA A100-SXM4-80GB<\/td>\n55.0GB<\/td>\n55.8GB<\/td>\n2.3 tok\/s<\/td>\n2026-09-30<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

All measurements and how we measure \u2192<\/a><\/p>\n

Video, images and audio we generated<\/h2>\n

We generate these ourselves from fixed prompts and sentences (no people, logos or reference audio) and show them in each article without picking or editing (videos are re-encoded to fit our upload limit). Each article also has the exact setup and run steps.<\/p>\n

\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n
Model<\/th>\nKind<\/th>\nFiles<\/th>\nWhere we ran it<\/th>\nGenerated<\/th>\n<\/tr>\n<\/thead>\n
LongLive-Plug-Wan2.2-TI2V-5B-cfg: Our Generated Video, File List<\/a><\/td>\nVideo<\/td>\n4 clips<\/td>\nCloud GPU (NVIDIA L40S)<\/td>\n2026-10-02<\/td>\n<\/tr>\n
LongLive-Plug-Wan2.1-T2V-14B-few-step: Our Generated Video, File List<\/a><\/td>\nVideo<\/td>\n4 clips<\/td>\nCloud GPU (NVIDIA L40S)<\/td>\n2026-10-02<\/td>\n<\/tr>\n
Supra2-IMG Image Generation Model: Our Generated Images, File List<\/a><\/td>\nImages<\/td>\n3 images<\/td>\nOur server’s CPU<\/td>\n2026-09-30<\/td>\n<\/tr>\n
Irodori-TTS-v4.1-Anime: Our Generated Audio, File List<\/a><\/td>\nSpeech<\/td>\n13 clips<\/td>\nOur server’s CPU<\/td>\n2026-09-30<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Everything we generated \u2192<\/a><\/p>\n

Latest by category<\/h2>\n

New Models<\/h3>\n
\n
\n
<\/div>\n
\n
New Models<\/span>2026-10-05<\/span><\/div>\n
Naive-N0.5-Flash MoE Model for AI Research and Coding: ~690GB Memory<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
New Models<\/span>2026-10-05<\/span><\/div>\n
Qwen3.8-Flash-Next Multimodal MoE Model: ~402GB Memory<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
New Models<\/span>2026-10-04<\/span><\/div>\n
LFM2.5-350M-Diffusion-Exp Text Generation Model: 4GB+ VRAM<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n

See all \u2192<\/a><\/p>\n

Image, Video and Audio<\/h3>\n
\n
\n
<\/div>\n
\n
Image, Video and Audio<\/span>2026-10-06<\/span><\/div>\n
Two Quantized Versions of Audio-Visual Video Generation Model FastH<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Image, Video and Audio<\/span>2026-10-06<\/span><\/div>\n
FastH3 Trim: 8-Step Text-to-Video Model with Audio Support<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Image, Video and Audio<\/span>2026-10-05<\/span><\/div>\n
DiarizationLM-Gemma-4-E4B-v1 Vision-Language Model: 8GB+ VRAM<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n

See all \u2192<\/a><\/p>\n

Engines and Tools<\/h3>\n
\n
\n
<\/div>\n
\n
Engines and Tools<\/span>2026-10-06<\/span><\/div>\n
magnitudedev\/magnitude v0.2.6 Released: Major Mac Speedups<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Engines and Tools<\/span>2026-10-06<\/span><\/div>\n
llama.cpp v0.6.0 Released with llama_batch_ext and Metal Boosts<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Engines and Tools<\/span>2026-10-06<\/span><\/div>\n
ggml v0.26.0 Released with Sparse Flash Attention & New Backends<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n

See all \u2192<\/a><\/p>\n

Community<\/h3>\n
\n
\n
<\/div>\n
\n
Community<\/span>2026-10-06<\/span><\/div>\n
Reflection Announces Beam: A 501B Open-Weight MoE Model<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Community<\/span>2026-10-04<\/span><\/div>\n
Aleph Alpha Releases Kolibri: A New Open-Weight MoE Model<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Community<\/span>2026-09-21<\/span><\/div>\n
Mini-AGI: A Byte-Level Continual Learning Model for 8GB VRAM<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n

See all \u2192<\/a><\/p>\n

Weekly Roundup<\/h3>\n
\n
\n
<\/div>\n
\n
Weekly Roundup<\/span>2026-10-05<\/span><\/div>\n
Local AI Model Weekly Highlights: Structured Output & Video LoRAs<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Weekly Roundup<\/span>2026-09-28<\/span><\/div>\n
Open-Weight Models Weekly: The MiMo-V2.6 Family & Transformers GGUF Support<\/a><\/div>\n<\/div>\n<\/div>\n
\n
<\/div>\n
\n
Weekly Roundup<\/span>2026-09-21<\/span><\/div>\n
Open-Weight Model Weekly: Ternary 27B Models and Apple Silicon Updates<\/a><\/div>\n<\/div>\n<\/div>\n<\/div>\n

See all \u2192<\/a><\/p>\n

Most read (past 28 days)<\/h2>\n
    \n
  1. \n
  2. \n
  3. \n
  4. \n
  5. \n
    \n
    5<\/div>\n
    MiMo-V2.6-Distill-Qwen-9B-GGUF: Our Test Answers, 12GB+ VRAM<\/span><\/a>
    New Models<\/span>2026-09-22 \u00b7 61 views<\/span><\/div>\n<\/div>\n<\/li>\n<\/ol>\n

    Last article published 2026-10-06 (JST).<\/em><\/p>\n","protected":false},"excerpt":{"rendered":"

    Local Model Watch is a reference site for open-weight generative models (text, image, video and audio) you can run on your own hardware, in English and Japanese: articles on new releases, hub pages by VRAM, model family, publisher, format and inference engine, and measurements we take ourselves by running models on our own server. Memory […]<\/p>\n","protected":false},"author":1,"featured_media":0,"parent":0,"menu_order":0,"comment_status":"closed","ping_status":"closed","template":"","meta":{"footnotes":""},"class_list":["post-1534","page","type-page","status-publish","hentry"],"lang":"en","translations":{"en":1534,"ja":1533},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1534","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages"}],"about":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/types\/page"}],"author":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/comments?post=1534"}],"version-history":[{"count":160,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1534\/revisions"}],"predecessor-version":[{"id":13618,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1534\/revisions\/13618"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=1534"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}