Conversation models that run on 24 GB

455 models with published weights that fit in 24 GB — a 24 GB card, or a Mac with 24. Counted at four-bit quantisation, weights only, leaving the machine about a third of its memory and a gigabyte for context: room for roughly 24 billion parameters. A mixture of experts is counted in full, because it is held in full even though only a few experts compute. 455 in all, a hundred to a page; this is page 5 of 5.

A 24 GB graphics card or a Mac configured with 24. This is the first size where the well-regarded mid-weight models — the twenty-something billion parameter class — run comfortably, and where a local model starts to be a real alternative to an API for daily work rather than a demonstration.

Models that answer in prose across whatever subject you put to them. This is the largest category and the least differentiated: most of them are competent at most things, so the choice usually comes down to price, context length and how the maker behaves about availability. Look at what a model costs per million tokens in and out — the output rate is often four or five times the input rate, and it is the one that decides your bill.

• bloom-560mBigScience Workshop0.6B≈0.4 GB at 4-bit
• Qwen1.5-0.5BAlibaba0.5B≈0.3 GB at 4-bit
• Qwen2-0.5BAlibaba0.5B≈0.3 GB at 4-bit
• reader-lm-0.5bJina AI0.5B≈0.3 GB at 4-bit
• Qwen2-0.5B-InstructAlibaba0.5B≈0.3 GB at 4-bit
• Qwen2.5-0.5BAlibaba0.5B≈0.3 GB at 4-bit
• Qwen2.5-0.5B-InstructAlibaba0.5B≈0.3 GB at 4-bit1 also selling it hosted
• blip-image-captioning-largeSalesforce0.5B≈0.3 GB at 4-bit
• LFM2.5-VL-450MLiquid AI0.4B≈0.3 GB at 4-bit
• PairRMLLM Blender0.4B≈0.3 GB at 4-bit
• CLIP-GmP-ViT-L-14zer0int0.4B≈0.3 GB at 4-bit
• circuit-sparsityOpenAI0.4B≈0.3 GB at 4-bit
• bart-large-cnnAI at Meta0.4B≈0.3 GB at 4-bit
• ModernBERT-largeAnswer.AI0.4B≈0.3 GB at 4-bit
• blip-vqa-baseSalesforce0.4B≈0.3 GB at 4-bit
• LFM2-350MLiquid AI0.4B≈0.2 GB at 4-bit
• LFM2.5-350MLiquid AI0.4B≈0.2 GB at 4-bit
• nougat-basefacebook0.3B≈0.2 GB at 4-bit
• bert-large-uncased-whole-word-masking-finetuned-squadBERT community0.3B≈0.2 GB at 4-bit
• mdeberta-v3-base-squad2timpal0l0.3B≈0.2 GB at 4-bit
• functiongemma-270m-itGoogle0.3B≈0.2 GB at 4-bit
• gemma-3-270mGoogle0.3B≈0.2 GB at 4-bit
• gemma-3-270m-itGoogle0.3B≈0.2 GB at 4-bit
• granite-docling-258MIBM Granite0.3B≈0.2 GB at 4-bit
• Florence-2-baseMicrosoft0.2B≈0.2 GB at 4-bit
• t5-baseT5 community0.2B≈0.1 GB at 4-bit
• chatgpt_paraphraser_on_T5_baseHumarin0.2B≈0.1 GB at 4-bit
• BiRefNetZhengPeng70.2B≈0.1 GB at 4-bit
• RMBG-2.0BRIA AI0.2B≈0.1 GB at 4-bit
• bert-base-multilingual-casedBERT community0.2B≈0.1 GB at 4-bit
• gpt-neo-125mEleutherAI0.2B≈0.1 GB at 4-bit
• ModernBERT-baseAnswer.AI0.1B≈0.1 GB at 4-bit
• MagicPrompt-Stable-DiffusionGustavosta0.1B≈0.1 GB at 4-bit
• gpt2OpenAI community0.1B≈0.1 GB at 4-bit
• distilbert-base-multilingual-caseddistilbert0.1B≈0.1 GB at 4-bit
• SmolLM2-135MHuggingFaceTB0.1B≈0.1 GB at 4-bit
• roberta-baseFacebook AI community0.1B≈0.1 GB at 4-bit
• roberta-base-squad2deepset0.1B≈0.1 GB at 4-bit
• bert-base-uncasedBERT community0.1B≈0.1 GB at 4-bit
• bert-base-casedBERT community0.1B≈0.1 GB at 4-bit
• bert-base-chineseBERT community0.1B≈0.1 GB at 4-bit
• BEN2Prama LLC0.1B≈0.1 GB at 4-bit
• distilgpt2DistilBERT community0.1B≈0.1 GB at 4-bit
• vit-base-patch16-224Google0.1B≈0.1 GB at 4-bit
• nsfw_image_detectionFalcons.ai0.1B≈0.1 GB at 4-bit
• distilbert-base-uncasedDistilBERT community0.1B≈0.0 GB at 4-bit
• t5-smallT5 community0.1B≈0.0 GB at 4-bit
• RMBG-1.4BRIA AI0.0B≈0.0 GB at 4-bit
• detr-resnet-50AI at Meta0.0B≈0.0 GB at 4-bit
• table-transformer-structure-recognitionMicrosoft0.0B≈0.0 GB at 4-bit
• table-transformer-detectionMicrosoft0.0B≈0.0 GB at 4-bit
• segformer_b2_clothesmattmdjaga0.0B≈0.0 GB at 4-bit
• segformer-b0-finetuned-ade-512-512NVIDIA0.0B≈0.0 GB at 4-bit
• Ornith-1.0-9BOrnith0.0B≈0.0 GB at 4-bit
• AutoGLM-Phone-9BZ.ai0.0B≈0.0 GB at 4-bit

Wider