Can I run it?
Pick your graphics card or machine and see which AI models run on it, and how fast.
Updated
Pick your graphics card or machine
Using a laptop, or a card or Mac that is not here? Find hardware checks every machine on our list.
Graphics cards
Macs and mini PCs
25 of 42 AI models run well on an RTX 3090 (24 GB of graphics memory). 17 don't fit.
An RTX 3090 costs $1,395€1,246, used (median of sold listings).
Runs well25
The whole model fits in the graphics memory. Bigger models come first: they are usually smarter, and a little slower.
- Nanbeige4.3-33B
- Qwen3.6-35B-A3B
- Qwen3 32B
- Qwen2.5 32B
- Gemma 4 31B
- Qwen3 Coder 30B A3B
- NVIDIA Nemotron 3.5 Lightning
- Gemma 2 27B
- Gemma 3 27B
- Gemma 4 26B A4B
- Qwen3.8-27B
- Muse Glimmer 30B
- OrcaCyber Nano 1.0
- Mistral Small 3.2 24B
- gpt-oss-20b
- DeepSeek R1 Distill Qwen 14B
- Phi-4 14B
- Qwen2.5 14B
- Gemma 2 9B
- NVIDIA Nemotron 3.5 Bolt E4B
- Llama 3.1 8B
- Mistral 7B v0.3
- Spark-X2.6-4B
- Llama 3.2 3B
- LFM3-1.7B
Doesn't fit17
Not enough graphics memory on the card alone.
- Recursive Flintneeds ~24 GB
- NVIDIA Nemotron 3.5 Bolt 74B-A9Bneeds ~45 GB
- DeepSeek-R1-Distill 70Bneeds ~46 GB
- Llama 3.3 70Bneeds ~46 GB
- Kolibri-1needs ~51 GB
- Qwen2.5 72Bneeds ~51 GB
- gpt-oss-120bneeds ~64 GB
- Llama 4 Scoutneeds ~68 GB
- NVIDIA Nemotron 3.5 Super VLneeds ~73 GB
- Ornith-1.5-122Bneeds ~77 GB
- Laguna S 2.1needs ~97 GB
- Qwen3.8-Flash-Nextneeds ~112 GB
- MiniMax M3.1 Flashneeds ~251 GB
- Reflection Beam 1.0needs ~308 GB
- DeepSeek V4.1 Flash 1017needs ~446 GB
- Meta Muse Spark Openneeds ~472 GB
- Kimi K3.1 Proneeds ~1681 GB
Every model on every card
The same answers in one table. Click a mark for the full answer, or a model for every card at once.
- Runs
- Runs slowly
- Doesn't fit
Swipe the table sideways to see all 10 cards.
Llama 3.3 70B on an RTX 4090 has its own written guide: Can I run Llama 3.3 70B on an RTX 4090?
Speeds and memory are our estimate from the hardware specs, not a measured test. Every answer is for the standard quality (Q4) and a conversation of about 12 pages of text (8,192 tokens). Prices excl. shipping unless marked. Euro prices incl. VAT where the seller charges it. Dollar prices are converted from our euro prices without the 21% EU VAT, before US sales tax. How we calculate
Browse by model or memory size
The ten largest models we track, a page for each memory size, and the calculator.
By model
- Can I run Kimi K3.1 Pro locally?
- Can I run Meta Muse Spark Open locally?
- Can I run DeepSeek V4.1 Flash 1017 locally?
- Can I run Reflection Beam 1.0 locally?
- Can I run MiniMax M3.1 Flash locally?
- Can I run Qwen3.8-Flash-Next locally?
- Can I run Ornith-1.5-122B locally?
- Can I run NVIDIA Nemotron 3.5 Super VL locally?
- Can I run Laguna S 2.1 locally?
- Can I run gpt-oss-120b locally?
By memory size
Questions
- How does localLLMrig calculate whether an LLM fits on my GPU or Mac?
- Every compatibility check on https://localllmrig.com/can-i-run adds three numbers for a 4-bit (Q4) model and an 8,192-token conversation (about 12 pages of text): (1) the model file size (a measured GGUF or MLX file where one is listed, or our formula estimate), (2) the conversation memory at 8,192 tokens (our estimate, using grouped-query attention and a sliding window where the model config records them), and (3) 0.80 GB of runtime working space (our estimate). A model is marked runs well when that total fits inside the machine's usable graphics memory.
- How much usable VRAM does a 12 GB, 16 GB, 24 GB, 32 GB, or 48 GB GPU actually have?
- Our calculator reserves 0.6 GB per graphics card for the operating system and display driver (our estimate), leaving 11.4 GB on a 12 GB card, 15.4 GB on a 16 GB card, 23.4 GB on a 24 GB card, 31.4 GB on a 32 GB card, and 46.8 GB across a two-card 48 GB setup (Dual RTX 3090, 128 GB RAM). On Apple Silicon, macOS shares unified memory with the system. The share in this catalog is 2/3 up to 32 GB and 75% above that: 10.7 GB on 16 GB (2/3), 21.3 GB on 32 GB (2/3); 48.0 GB on 64 GB (75%), 72.0 GB on 96 GB (75%), 96.0 GB on 128 GB (75%), 192.0 GB on 256 GB (75%).
- How many open-weight models fit on a 24 GB GPU like the RTX 3090 or RTX 4090?
- As of 24 Oct 2026, 25 of the 42 AI models we track on https://localllmrig.com/can-i-run run well at standard 4-bit quality (Q4, 8,192 tokens) inside the 23.4 GB usable memory of a 24 GB card, including Qwen3 32B (https://localllmrig.com/can-i-run/qwen3-32b, 22.7 GB est.), Qwen3.6-35B-A3B (https://localllmrig.com/can-i-run/qwen3.6-35b-a3b, 23.1 GB est.), Gemma 4 31B (https://localllmrig.com/can-i-run/gemma-4-31b, 20.6 GB est.), NVIDIA Nemotron 3.5 Lightning (https://localllmrig.com/can-i-run/nemotron-3.5-lightning, 19.8 GB est.), Gemma 4 26B A4B (https://localllmrig.com/can-i-run/gemma-4-26b-a4b, 18.1 GB est.), Qwen3.8-27B (https://localllmrig.com/can-i-run/qwen3.8-27b, 17.8 GB est.). In addition, Recursive Flint (https://localllmrig.com/can-i-run/flint) needs 24.5 GB at its standard file (1.1 GB over a 24 GB card) and fits that card at UD-Q3_K_XL (18.49 GB measured GGUF, 19.6 GB total estimate).
- Can I test these open models live in my browser before downloading 17 to 96 GB of weights?
- Yes. On https://localllmrig.com/, you can test 10 open-weight models (26B to 280B parameters) side by side in your browser for free with no account, including Gemma 4 31B (https://localllmrig.com/can-i-run/gemma-4-31b), Gemma 4 26B A4B (https://localllmrig.com/can-i-run/gemma-4-26b-a4b), NVIDIA Nemotron 3.5 Lightning (https://localllmrig.com/can-i-run/nemotron-3.5-lightning), Laguna S 2.1 (https://localllmrig.com/can-i-run/laguna-s-2.1), NVIDIA Nemotron 3.5 Super VL (https://localllmrig.com/can-i-run/nemotron-3.5-super-vl), Recursive Flint (https://localllmrig.com/can-i-run/flint). Each of those model pages also includes an inline browser test.
- Are the tokens-per-second (tok/s) speeds and hardware prices measured?
- Speeds (tok/s) and total memory figures in the compatibility tables are our estimates from hardware and model specifications, not measured llama-bench runs. Hardware prices cite the dated source stored on each machine. Used sold-listing medians in this catalog: GeForce RTX 3060 12GB at €261 (pcprice.watch); GeForce RTX 3090 24GB at €1,246 (pcprice.watch); GeForce RTX 4090 24GB at €2,185 (pcprice.watch). Other source names on the hardware list include pcprice.watch, GPU PRIX, PriceSquirrel, Belgian Apple Store, ASUS eStore Belgium, NVIDIA, idealo.de, Framework (Germany), GEEKOM (Germany), Minisforum EU store, Beelink EU store, Intel. A machine with no checked listing is marked price unverified.