{"id":1202,"date":"2026-09-17T06:20:39","date_gmt":"2026-09-16T21:20:39","guid":{"rendered":"https:\/\/localmodelwatch.tsuchitsuchi.com\/vram-guide-en\/"},"modified":"2026-09-19T06:50:53","modified_gmt":"2026-09-18T21:50:53","slug":"vram-guide-en","status":"publish","type":"page","link":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/vram-guide-en\/","title":{"rendered":"Local Models by VRAM: Quick Reference"},"content":{"rendered":"

Open-weight models covered by Local Model Watch, grouped by the smallest amount of GPU memory (VRAM) they are estimated to run in. Each figure is computed by this site from the actual size of the distributed files plus a runtime overhead \u2014 not quoted from model cards. See our editorial policy<\/a> for the method.<\/p>\n

Models listed under a tier do not<\/strong> fit in the smaller tiers. If you have a 24GB GPU, the 8GB, 12GB, 16GB and 24GB tables all apply to you. Models that don’t fit even the largest tier (80GB) are listed separately at the bottom, since that too is something readers need to know.<\/p>\n

Last updated 2026-09-19 (JST). 28 models listed.<\/em><\/p>\n

4GB (laptop iGPU \/ phone class)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
Cactus Compute\u30018\u301c29MB\u306e\u81ea\u52d5\u5316\u30e2\u30c7\u30eb\u300cNeedle 3\u300d\u516c\u958b<\/td>\n–<\/td>\n\u305d\u306e\u307e\u307e\u306e\u7cbe\u5ea6 (0.2GB)<\/td>\n0.3GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
Lightricks\/LTX-2.5-22b-IC-LoRA-Ingredients<\/td>\n–<\/td>\n\u305d\u306e\u307e\u307e\u306e\u7cbe\u5ea6 (1.2GB)<\/td>\n1.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
openbmb\/MiniCPM5-2B<\/td>\n2.5B<\/td>\nQ8_0 (2.5GB)<\/td>\n3.0GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

8GB (RTX 4060 \/ 3060 Ti, etc.)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
prism-ml\/Ternary-Bonsai-2-27B-gguf<\/td>\n27.8B<\/td>\nQ1_0 (5.5GB)<\/td>\n6.6GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
harshatheg\/Qwen-2.5-1B-RLCD<\/td>\n1.5B<\/td>\nBF16 (2.9GB)<\/td>\n3.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
EleutherAI\/bergson-wikitext-gpt2-leaderboard<\/td>\n–<\/td>\n\u305d\u306e\u307e\u307e\u306e\u7cbe\u5ea6 (6.5GB)<\/td>\n7.8GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
bartowski\/TheDrummer_Orion-26B-A4B-v1.1-GGUF<\/td>\n25.8B<\/td>\nF16 (1.1GB)<\/td>\n1.3GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
openbmb\/MiniCPM5-2B-GGUF<\/td>\n–<\/td>\nF16 (4.7GB)<\/td>\n5.6GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

12GB (RTX 4070 \/ 3060 12GB, etc.)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
tencent\/WeVisDoc-4B<\/td>\n4.4B<\/td>\nBF16 (9.0GB)<\/td>\n10.8GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
prism-ml\/Ternary-Bonsai-2-27B-mlx-2bit<\/td>\n27.4B<\/td>\nMLX 2bit (8.0GB)<\/td>\n9.6GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
prism-ml\/Ternary-Bonsai-2-27B-gguf-dev<\/td>\n27.8B<\/td>\nQ2_0 (7.1GB)<\/td>\n8.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
Mothersuperior\/yue2-mothersuperior-realaudio-tokenizer-v4<\/td>\n3.6B<\/td>\nBF16 (6.8GB)<\/td>\n8.1GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
bartowski\/vectionlabs_Salience-27B-R6-GGUF<\/td>\n27.8B<\/td>\nIQ2_M (9.8GB)<\/td>\n11.8GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
tencent\/Simple-Attention-Sparsification<\/td>\n4.0B<\/td>\nBF16 (7.5GB)<\/td>\n9.0GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
Comfy-Org\/YuE2<\/td>\n3.6B<\/td>\nBF16 (6.8GB)<\/td>\n8.1GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
bartowski\/Gryphe_Pantheon-Reasoning-26B-A4B-1.1-V2-GGUF<\/td>\n26.5B<\/td>\nIQ2_S (10.0GB)<\/td>\n12.0GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
m-a-p\/YuE2-3B<\/td>\n3.6B<\/td>\nBF16 (6.8GB)<\/td>\n8.1GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

16GB (RTX 5060 Ti 16GB \/ 4060 Ti 16GB, etc.)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
bartowski\/nex-agi_Nex-N2.5-mini-GGUF<\/td>\n35.1B<\/td>\nQ2_K (12.9GB)<\/td>\n15.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
agentionai\/Signal-3.8-27B-GGUF<\/td>\n27.8B<\/td>\nIQ4_XS (13.3GB)<\/td>\n15.9GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
nex-agi\/Nex-N2.5-mini<\/td>\n35.1B<\/td>\nQ2_K (12.9GB)<\/td>\n15.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
Jackrong\/Qwopus3.8-27B-Flash-GGUF<\/td>\n27.8B<\/td>\nQ3_K_M (12.6GB)<\/td>\n15.1GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

24GB (RTX 4090 \/ 3090, etc.)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
EleutherAI\/olmo3-7b-sdf-sft-clean150<\/td>\n7.3B<\/td>\nBF16 (13.6GB)<\/td>\n16.3GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

80GB class (A100 \/ H100)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
Edge0\/Edge0-35B-A3B-preview<\/td>\n34.7B<\/td>\nU32 (64.6GB)<\/td>\n77.5GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n

Over 80GB (datacenter GPU \/ multi-GPU required)<\/h2>\n
\n

\u2192 Scroll horizontally to see all columns<\/p>\n\n\n\n\n\n\n\n\n\n
Model<\/th>\nParameters<\/th>\nSmallest build<\/th>\nEst. memory<\/th>\nArticle<\/th>\n<\/tr>\n<\/thead>\n
bartowski\/Intern-S2-397B-GGUF<\/td>\n403.4B<\/td>\nIQ1_S (85.3GB)<\/td>\n102.4GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
dealignai\/DeepSeek-V4.1-Flash-UNCENSORED-FP8<\/td>\n763.2B<\/td>\nBF16 (902.8GB)<\/td>\n1083.3GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
deepseek-ai\/DeepSeek-V4.1-Flash<\/td>\n763.2B<\/td>\nBF16 (902.8GB)<\/td>\n1083.3GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
nex-agi\/Nex-N2.5-Pro<\/td>\n396.8B<\/td>\nIQ1_S (81.8GB)<\/td>\n98.2GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n
IFM\/K2-Horizon-MoVA-36B-A4B-GGUF<\/td>\n37.4B<\/td>\nBF16 (69.8GB)<\/td>\n83.7GB<\/td>\nRead<\/a><\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Open-weight models covered by Local Model Watch, grouped by the smallest amount of GPU memory (VRAM) they are estimated to run in. Each figure is computed by this site from the actual size of the distributed files plus a runtime overhead \u2014 not quoted from model cards. See our editorial policy for the method. Models […]<\/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-1202","page","type-page","status-publish","hentry"],"lang":"en","translations":{"en":1202,"ja":1200},"_links":{"self":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1202","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=1202"}],"version-history":[{"count":10,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1202\/revisions"}],"predecessor-version":[{"id":1823,"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/pages\/1202\/revisions\/1823"}],"wp:attachment":[{"href":"https:\/\/localmodelwatch.tsuchitsuchi.com\/en\/wp-json\/wp\/v2\/media?parent=1202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}