Models you can run at home
Each page shows the sizes, how much memory each one needs at different conversation lengths, which computers run it, and how to start it.
Text, chat and coding
31A leading 2026 all-round model that fits a 24 GB graphics card. Reads images and video too.
An experimental preview of the next Qwen design, and the most-used model on 128 GB machines. Smaller 3-bit and 4-bit files fit 96 GB.
April 2026 Qwen focused on coding agents: a fast 35B-A3B and a 27B dense, both reading images.
A full family from phone size (0.8B) to 122B, every size reading images. A strong pick for small machines.
The 2025 Qwen text models. Older now, but supported everywhere and well tested.
Qwen models made for coding agents. The 30B-A3B fits a 24 GB card and answers fast.
Qwen’s 2025 vision models for photos, documents and screen control.
Google’s 2026 family, from phone size to 31B. Reads images and works in 140+ languages.
OpenAI’s open models for reasoning and tool use. The 20B fits a 16 GB card and is fast.
Meta’s 2024 models. Older and weaker than 2026 models, but supported by every tool.
Meta’s August 2026 open agent model, built to run tool-using tasks on one consumer graphics card.
Microsoft’s small models trained on textbook-style data. Good at maths for their size.
Mistral’s small models for laptops and 12–16 GB cards, with image input and a 256K context.
Mistral’s 24B coding-agent model. Mistral says it runs on one RTX 4090 or a 32 GB Mac.
A 119B mixture-of-experts model that merges Mistral’s chat, reasoning and coding lines. For 96–128 GB machines.
Mistral’s 128B dense flagship. Fits 96–128 GB machines, but every word uses all 128B parameters, so it is slow.
IBM’s business-focused models (3B, 8B, 30B) with a thinking mode and tool calls.
NVIDIA reasoning models with a hybrid design that keeps conversation memory tiny, even at long context.
NVIDIA’s August 2026 30B-A3B, released with its training data under a permissive licence.
The Z.ai models you can run at home: a 30B-A3B for 24 GB machines and a 106B for 64–96 GB.
Coding-agent models built on Qwen3.5 and Gemma 4, in a 9B and a fast 35B-A3B.
Poolside’s coding-agent models: XS (33B-A3B) for 24–36 GB machines, S (118B) for 128 GB.
Cohere’s 30B-A3B research model for code and terminal tasks.
Xiaomi’s small local edition of MiMo-V2.6: Qwen3.5 9B trained further on agent and coding data.
Tiny 1B and 2B models for phones and old laptops, with open training data.
Liquid AI’s small on-device models, tuned for tool use rather than coding or knowledge.
A fully open 3B model from Hugging Face: weights, data and training recipe all published.
Ai2’s fully open models (7B and 32B), each in a chat and a step-by-step thinking version.
Moonshot’s long-context model: 1M tokens with a tiny conversation memory.
Z.ai’s 320B model that reads images, with a 1M context. The main pick for 192 to 512 GB machines.
DeepSeek’s 284B model with a 1M context and very small conversation memory. Fits 192 GB at 4-bit and 128 GB at 3-bit.
Image generation
6A 6B image model that makes a picture in 8 steps and fits a 16 GB card. Apache 2.0.
Black Forest Labs image models. The 4B “klein” is small, fast and free for business; the 9B and the big “dev” are non-commercial.
Alibaba’s image models, strong at drawing readable text. 2512 is free for business; 2.1 is research-only.
An 8B image model that works directly on pixels: generation, editing and personalised subjects. MIT licence.
A 12B image model distilled to 8 steps. Free for business only under $1M yearly revenue.
Stability AI’s image models. SDXL is old but small and still the most downloaded; 3.5 Medium is newer.
Text to speech
7A tiny 82M voice model with 54 preset voices in 8 languages. Runs on almost anything, even without a graphics card.
Alibaba’s voice models: clone a voice from 3 seconds, preset voices, or design a voice from a description. 10 languages.
Resemble AI’s voice-cloning models: Multilingual V3 covers 23 languages; Turbo is English-only and quicker.
A 2B voice model for 30 languages with high-quality 48 kHz audio, voice design and cloning.
A 0.6B voice model covering 600+ languages with cloning. The weights are non-commercial.
A 100M voice model for 20 languages that its authors say runs on a 4-core CPU.
A small on-device voice model for 31 languages that runs without a graphics card.
Speech to text
7OpenAI’s speech-to-text model for 99 languages, fast through faster-whisper. The default choice for transcription.
NVIDIA’s fast transcription model for 25 European languages, with word timings.
Alibaba’s speech recognition for 30 languages and 22 Chinese dialects, in 1.7B and 0.6B sizes.
NVIDIA’s low-delay streaming speech recognition for voice agents, 40 language-locales.
Mistral’s live transcription model for 13 languages with about half a second of delay.
Cohere’s 2B speech recognition for 14 languages, splitting long audio automatically.
Microsoft’s 8.7B model that transcribes up to an hour in one pass, with who-said-what and timings.