# FitMyModel

> FitMyModel matches popular open AI models (Llama, Qwen, Gemma, gpt-oss, DeepSeek, FLUX, Whisper) to new-retail PCs, mini PCs and parts that can run them at home. Pages are compiled from cited sources and official docs; we do not test hardware ourselves. Prices show the date and time they were checked.

## Core
- [Model finder](https://fitmymodel.com/models/): memory needed by leaderboard-ranked open models and the smallest hardware tier for each
- [Hardware tiers](https://fitmymodel.com/tiers/): five memory tiers from 8–12 GB GPUs to 128 GB unified memory

- [What fits my memory](https://fitmymodel.com/what-fits/): pick 8–128 GB and see which ranked models fit
- [Can it run?](https://fitmymodel.com/run/): 48 model-on-device pages (the hub lists all) with fit, quantization and the Ollama command

## By memory size
- [What local AI models run on 8 GB of VRAM?](https://fitmymodel.com/vram/8gb/): 1 ranked models fit
- [What local AI models run on 12 GB of VRAM?](https://fitmymodel.com/vram/12gb/): 2 ranked models fit
- [What local AI models run on 16 GB of VRAM?](https://fitmymodel.com/vram/16gb/): 5 ranked models fit
- [What local AI models run on 24 GB of VRAM?](https://fitmymodel.com/vram/24gb/): 9 ranked models fit
- [What local AI models run on 32 GB of VRAM?](https://fitmymodel.com/vram/32gb/): 9 ranked models fit
- [What local AI models run on 48 GB of VRAM?](https://fitmymodel.com/vram/48gb/): 10 ranked models fit
- [What local AI models run on 64 GB of VRAM?](https://fitmymodel.com/vram/64gb/): 10 ranked models fit
- [What local AI models run on 96 GB of VRAM?](https://fitmymodel.com/vram/96gb/): 13 ranked models fit
- [What local AI models run on 128 GB of VRAM?](https://fitmymodel.com/vram/128gb/): 13 ranked models fit

## By card or device
- [Intel Arc B580 12GB](https://fitmymodel.com/tiers/gpu-8-12gb/): 12 GB for the GPU; runs Phi-4 14B, Llama 3.1 8B
- [NVIDIA GeForce RTX 5060 Ti 16GB](https://fitmymodel.com/tiers/gpu-16gb/): 16 GB for the GPU; runs Gemma 4 26B-A4B (MoE), Mistral Small 3.2 24B, gpt-oss 20B, Phi-4 14B, Llama 3.1 8B
- [Intel Arc Pro B60 24GB](https://fitmymodel.com/tiers/gpu-24-32gb/): 24 GB for the GPU; runs Gemma 4 31B, Gemma 4 26B-A4B (MoE), Qwen3.8 27B, Qwen3.5 27B, Qwen3.5 35B-A3B (MoE), Mistral Small 3.2 24B, gpt-oss 20B, Phi-4 14B, Llama 3.1 8B
- [AMD Radeon AI Pro R9700 32GB](https://fitmymodel.com/tiers/gpu-24-32gb/): 32 GB for the GPU; runs Gemma 4 31B, Gemma 4 26B-A4B (MoE), Qwen3.8 27B, Qwen3.5 27B, Qwen3.5 35B-A3B (MoE), Mistral Small 3.2 24B, gpt-oss 20B, Phi-4 14B, Llama 3.1 8B
- [Ryzen AI Max+ 395 mini PC, 64 GB](https://fitmymodel.com/tiers/unified-64gb/): 48 GB for the GPU; runs Gemma 4 31B, Gemma 4 26B-A4B (MoE), Qwen3.8 27B, Qwen3.5 27B, Qwen3.5 35B-A3B (MoE), Mistral Small 3.2 24B, Llama 3.3 70B, gpt-oss 20B, Phi-4 14B, Llama 3.1 8B
- [Ryzen AI Max+ 395 mini PC, 128 GB](https://fitmymodel.com/tiers/unified-128gb/): 96 GB for the GPU; runs Gemma 4 31B, Gemma 4 26B-A4B (MoE), Qwen3.8 27B, Qwen3.5 122B-A10B (MoE), Nemotron 3.5 Super 120B-A12B, Qwen3.5 27B, Qwen3.5 35B-A3B (MoE), Mistral Small 3.2 24B, gpt-oss 120B, Llama 3.3 70B, gpt-oss 20B, Phi-4 14B, Llama 3.1 8B

## By model
- [Gemma 4 31B](https://fitmymodel.com/guides/what-runs-gemma-4-31b/): needs 19–20 GB (Q4_K_M); pages: https://fitmymodel.com/run/gemma-4-31b-on-radeon-ai-pro-r9700/ https://fitmymodel.com/run/gemma-4-31b-on-strix-halo-64gb/
- [Gemma 4 26B-A4B (MoE)](https://fitmymodel.com/models/): needs 16 GB (default tag); pages: https://fitmymodel.com/run/gemma-4-26b-a4b-on-rtx-5060-ti-16gb/ https://fitmymodel.com/run/gemma-4-26b-a4b-on-arc-pro-b60/
- [Qwen3.8 27B](https://fitmymodel.com/models/): needs 18 GB (Q4_K_M); pages: https://fitmymodel.com/run/qwen3-8-27b-on-arc-pro-b60/ https://fitmymodel.com/run/qwen3-8-27b-on-strix-halo-128gb/
- [Qwen3.5 122B-A10B (MoE)](https://fitmymodel.com/guides/what-runs-qwen3-5/): needs 81 GB (Q4_K_M); pages: https://fitmymodel.com/run/qwen3-5-122b-a10b-on-strix-halo-128gb/
- [Nemotron 3.5 Super 120B-A12B](https://fitmymodel.com/models/): needs 90 GB (Q4_K_M (default tag 80 GB)); pages: https://fitmymodel.com/run/nemotron-3-5-super-on-strix-halo-128gb/
- [Qwen3.5 27B](https://fitmymodel.com/guides/what-runs-qwen3-5/): needs 17–20 GB (Q4_K_M); pages: see the /run/ hub
- [Qwen3.5 35B-A3B (MoE)](https://fitmymodel.com/guides/what-runs-qwen3-5/): needs 22 GB (Q4_K_M); pages: https://fitmymodel.com/run/qwen3-5-35b-a3b-on-arc-pro-b60/
- [Mistral Small 3.2 24B](https://fitmymodel.com/models/): needs 15 GB (Q4_K_M); pages: https://fitmymodel.com/run/mistral-small-3-2-on-rtx-5060-ti-16gb/
- [gpt-oss 120B](https://fitmymodel.com/guides/what-runs-gpt-oss-120b/): needs 65 GB (MXFP4 (as shipped)); pages: https://fitmymodel.com/run/gpt-oss-120b-on-strix-halo-128gb/
- [Llama 3.3 70B](https://fitmymodel.com/guides/what-runs-llama-3-3-70b/): needs 43 GB (Q4_K_M); pages: https://fitmymodel.com/run/llama-3-3-70b-on-strix-halo-64gb/
- [gpt-oss 20B](https://fitmymodel.com/guides/what-runs-gpt-oss-20b/): needs 14 GB (MXFP4 (as shipped)); pages: https://fitmymodel.com/run/gpt-oss-20b-on-rtx-5060-ti-16gb/ https://fitmymodel.com/run/gpt-oss-20b-on-radeon-ai-pro-r9700/
- [Phi-4 14B](https://fitmymodel.com/guides/what-runs-llama-3-1-8b-phi-4-qwen3-14b/): needs 9.1 GB (Q4_K_M); pages: https://fitmymodel.com/run/phi-4-on-arc-b580/ https://fitmymodel.com/run/phi-4-on-rtx-5060-ti-16gb/
- [Llama 3.1 8B](https://fitmymodel.com/guides/what-runs-llama-3-1-8b-phi-4-qwen3-14b/): needs 4.9 GB (Q4_K_M); pages: https://fitmymodel.com/run/llama-3-1-8b-on-arc-b580/ https://fitmymodel.com/run/llama-3-1-8b-on-rtx-5060-ti-16gb/

## By task
- [Chat and assistants: What runs gpt-oss 20B? PCs and builds that fit](https://fitmymodel.com/guides/what-runs-gpt-oss-20b/): gpt-oss 20B is a 14 GB download in Ollama. See the 16 GB and 24 GB GPUs, prebuilts and parts that fit it, with prices checked Oct 27, 2026.
- [Chat and assistants: What runs gpt-oss 120B? 128 GB unified-memory mini PCs](https://fitmymodel.com/guides/what-runs-gpt-oss-120b/): gpt-oss 120B is a 65 GB download in Ollama, so it needs a 128 GB unified-memory box. Compare the Strix Halo mini PCs that fit it.
- [Chat and assistants: What runs Llama 3.3 70B? Unified-memory mini PCs vs two GPUs](https://fitmymodel.com/guides/what-runs-llama-3-3-70b/): Llama 3.3 70B is a 43 GB download at Q4_K_M. Compare 64 GB and 128 GB unified-memory mini PCs with a two-GPU build.
- [Chat and assistants: What runs Qwen3.5? Hardware for 27B, 35B-A3B and 122B-A10B](https://fitmymodel.com/guides/what-runs-qwen3-5/): Qwen3.5 27B needs 17–20 GB, 35B-A3B needs 22 GB and 122B-A10B needs 81 GB in Ollama. Here is the hardware tier for each size.
- [Chat and assistants: What runs Gemma 4 31B? 24–32 GB GPUs and mini PCs](https://fitmymodel.com/guides/what-runs-gemma-4-31b/): Gemma 4 31B is a 19–20 GB download in Ollama with a 256K context window. See the GPUs and systems that hold it.
- [Chat and assistants: What runs DeepSeek-R1 32B? GPUs and systems that fit](https://fitmymodel.com/guides/what-runs-deepseek-r1-32b/): The DeepSeek-R1 32B distill is a 20 GB download in Ollama. See the 24–32 GB GPUs and unified-memory boxes that run it.
- [Chat and assistants: What runs Llama 3.1 8B, Phi-4 and Qwen3 14B? Entry GPUs](https://fitmymodel.com/guides/what-runs-llama-3-1-8b-phi-4-qwen3-14b/): Llama 3.1 8B (4.9 GB), Phi-4 (9.1 GB) and Qwen3 14B (9.3 GB) fit 12 GB GPUs. See the entry cards and systems that run them.
- [Image generation: Run FLUX and SDXL at home: which GPU do you need?](https://fitmymodel.com/guides/flux-sdxl-which-gpu/): SDXL runs on 8 GB of VRAM per Stability AI; FLUX.1 dev needs about 12 GB at FP8 (third-party estimate). Pick the GPU tier for image generation.
- [Transcription: Whisper on your own PC: hardware for transcription](https://fitmymodel.com/guides/whisper-hardware/): OpenAI lists about 10 GB of VRAM for Whisper large and about 6 GB for turbo. See the GPUs that run local transcription.
- [Clustering and storage: Clustering and NAS](https://fitmymodel.com/clustering-nas/): joining machines and storing models

## Tiers
- [8–12 GB GPU](https://fitmymodel.com/tiers/gpu-8-12gb/): 8B–14B chat models at Q4, SDXL, Whisper, FLUX at Q4
- [16 GB GPU](https://fitmymodel.com/tiers/gpu-16gb/): gpt-oss 20B, Mistral Small 3.2 24B (tight) and FLUX at FP8, plus everything in the 8–12 GB tier
- [24–32 GB GPU](https://fitmymodel.com/tiers/gpu-24-32gb/): 27B–35B models at Q4 (Qwen3.5 27B, Gemma 4 31B, DeepSeek-R1 32B)
- [Unified 64 GB](https://fitmymodel.com/tiers/unified-64gb/): 35B MoE models; Llama 3.3 70B at Q4 (tight)
- [Unified 128 GB](https://fitmymodel.com/tiers/unified-128gb/): gpt-oss 120B, Qwen3.5 122B-A10B, Llama 3.3 70B with room

## Series: Why run AI locally
- [Why run AI locally (series hub)](https://fitmymodel.com/why-local/): advantages of local models: privacy, cost, offline use, control, and honest downsides
- [Part 1: Where do your AI chats go, and does a local model keep them private?](https://fitmymodel.com/why-local/where-do-your-ai-chats-go/): Cloud chatbots keep your chats for 72 hours to 5 years depending on the service and your settings. A local model keeps them on your own computer.
- [Part 2: Is local AI cheaper than a ChatGPT or Claude subscription?](https://fitmymodel.com/why-local/is-local-ai-cheaper-than-a-subscription/): ChatGPT Plus, Claude Pro and Google AI Pro cost about $20 a month. A $607 mini PC equals 30 months of that. Here is the math, and why it isn’t a like-for-like trade.
- [Part 3: Can you use AI with no internet connection?](https://fitmymodel.com/why-local/can-you-use-ai-with-no-internet/): Yes, with a local model. You need the internet once to install the app and download a model; after that, chatting runs entirely on your computer.
- [Part 4: Why do cloud AI limits and rules keep changing, and does local AI avoid that?](https://fitmymodel.com/why-local/cloud-ai-limits-and-rule-changes/): Cloud plans cap how much you can use and retire models on the provider’s schedule. A model on your own drive has no usage cap and stays the same until you change it.
Planned parts: Is local AI faster than ChatGPT? Latency explained; How do you customize a local AI model?; Will your AI tools still work in five years?; Is local AI a good fit for families and small businesses?; When is cloud AI still better than running it locally?

## Articles and news
- [Articles index](https://fitmymodel.com/articles/): explainers, buying advice and monthly news
- [Can a $600 mini PC really run local AI?](https://fitmymodel.com/articles/can-a-600-dollar-mini-pc-run-local-ai/): Yes, for small models, if you buy enough RAM. Here is what fits in a $607 Minisforum box, and what it can’t do.
- [Why VRAM matters more than GPU speed for LLMs](https://fitmymodel.com/articles/why-vram-matters-more-than-gpu-speed/): A faster GPU that can’t hold the model loses to a slower one that can. Here’s the one spec to read first.
- [Unified memory vs a GPU: Strix Halo, Mac and RTX compared](https://fitmymodel.com/articles/unified-memory-vs-gpu-strix-halo-mac-rtx/): Capacity or bandwidth? Three ways to hold a model, compared on the two numbers that matter, straight from the spec sheets.
- [Is 16 GB enough for local AI in 2026?](https://fitmymodel.com/articles/is-16gb-enough-in-2026/): 16 GB of VRAM is the new sweet spot. 16 GB of system RAM alone is the bare minimum. Here’s where the line falls.
- [Running AI privately at home: what you give up and what you gain](https://fitmymodel.com/articles/running-ai-privately-at-home/): Your prompts never leave the house. Here’s the honest trade: privacy and no subscriptions, in exchange for smaller models and a bigger upfront bill.
- [Clustering two machines for bigger models: worth it?](https://fitmymodel.com/articles/clustering-two-machines-for-bigger-models/): Two boxes can hold a model neither could alone. But the tools are early, and the network is now part of your AI.
- [The cheapest way to run gpt-oss 20B](https://fitmymodel.com/articles/cheapest-way-to-run-gpt-oss-20b/): OpenAI built it for 16 GB. Here are the three cheapest routes we priced, from a $607 mini PC to a complete RTX PC.
- [New open models this month (October 2026), and the hardware they need](https://fitmymodel.com/articles/new-open-models-october-2026/): Four real releases, from a 1 GB edge model to an 816B giant. Here’s which ones fit a home machine.

## Model guides
- [What runs gpt-oss 20B? PCs and builds that fit](https://fitmymodel.com/guides/what-runs-gpt-oss-20b/): gpt-oss 20B is a 14 GB download in Ollama. See the 16 GB and 24 GB GPUs, prebuilts and parts that fit it, with prices checked Oct 27, 2026.
- [What runs gpt-oss 120B? 128 GB unified-memory mini PCs](https://fitmymodel.com/guides/what-runs-gpt-oss-120b/): gpt-oss 120B is a 65 GB download in Ollama, so it needs a 128 GB unified-memory box. Compare the Strix Halo mini PCs that fit it.
- [What runs Llama 3.3 70B? Unified-memory mini PCs vs two GPUs](https://fitmymodel.com/guides/what-runs-llama-3-3-70b/): Llama 3.3 70B is a 43 GB download at Q4_K_M. Compare 64 GB and 128 GB unified-memory mini PCs with a two-GPU build.
- [What runs Qwen3.5? Hardware for 27B, 35B-A3B and 122B-A10B](https://fitmymodel.com/guides/what-runs-qwen3-5/): Qwen3.5 27B needs 17–20 GB, 35B-A3B needs 22 GB and 122B-A10B needs 81 GB in Ollama. Here is the hardware tier for each size.
- [What runs Gemma 4 31B? 24–32 GB GPUs and mini PCs](https://fitmymodel.com/guides/what-runs-gemma-4-31b/): Gemma 4 31B is a 19–20 GB download in Ollama with a 256K context window. See the GPUs and systems that hold it.
- [What runs DeepSeek-R1 32B? GPUs and systems that fit](https://fitmymodel.com/guides/what-runs-deepseek-r1-32b/): The DeepSeek-R1 32B distill is a 20 GB download in Ollama. See the 24–32 GB GPUs and unified-memory boxes that run it.
- [What runs Llama 3.1 8B, Phi-4 and Qwen3 14B? Entry GPUs](https://fitmymodel.com/guides/what-runs-llama-3-1-8b-phi-4-qwen3-14b/): Llama 3.1 8B (4.9 GB), Phi-4 (9.1 GB) and Qwen3 14B (9.3 GB) fit 12 GB GPUs. See the entry cards and systems that run them.
- [Run FLUX and SDXL at home: which GPU do you need?](https://fitmymodel.com/guides/flux-sdxl-which-gpu/): SDXL runs on 8 GB of VRAM per Stability AI; FLUX.1 dev needs about 12 GB at FP8 (third-party estimate). Pick the GPU tier for image generation.
- [Whisper on your own PC: hardware for transcription](https://fitmymodel.com/guides/whisper-hardware/): OpenAI lists about 10 GB of VRAM for Whisper large and about 6 GB for turbo. See the GPUs that run local transcription.

## Other
- [10 GbE for local AI: link two boxes and a NAS for models](https://fitmymodel.com/clustering-nas/): Networking for local AI only: 10 GbE switches and cards, USB4 links, a NAS for model files, and the llama.cpp RPC backend as a labeled experiment.
- [How we pick hardware](https://fitmymodel.com/how-we-test/): How FitMyModel picks hardware, how models get a tier and how install steps are sourced.
- [Minisforum MS-S1 MAX 128GB at $3,919: runs gpt-oss 120B and Qwen3.5 122B](https://fitmymodel.com/deals/2026-10-27-minisforum-ms-s1-max-128gb-3919/): Price watch: the 128 GB / 2 TB MS-S1 MAX was $3,919 on the Minisforum store at 12:40 PM CT on Oct 27, 2026. Which models fit, and the caveats.
- [ASRock Arc B580 12GB at $329.99: runs 8B–14B models and SDXL](https://fitmymodel.com/deals/2026-10-27-asrock-arc-b580-12gb-329/): Price watch: the ASRock Challenger Arc B580 12GB was $329.99, sold by Newegg, on Oct 27, 2026. Which models fit a 12 GB card, and the caveats.
- [Affiliate disclosure](https://fitmymodel.com/affiliate-disclosure/): how the site earns money
- [Contact](https://fitmymodel.com/contact/): how to reach the site
- [Privacy policy](https://fitmymodel.com/privacy/): what data the site collects

Full text: https://fitmymodel.com/llms-full.txt