US prices · checked 0 min ago
Plan a
local AI machine
See which GPUs, Macs and mini PCs run the models you want, how fast they run them, and what they cost at retailers in the US today.
Featured builds
Ranked by the best model each one runs fast enough for coding, at 32k context.
- 1
Mini PC · 128 GB · 256 GB/s
Qwen 3.5 122B-A10BUD-Q2_K_XL
20tok/sest.
Memory40.8 / 96.0 GB$1,600
- 2
PC · 16 GB · 448 GB/s
gpt-oss 20BQ4_K_M
53tok/sest.
Memory12.5 / 15.4 GB$1,687
- 3
PC · 12 GB · 360 GB/s
Gemma 4 12BQ6_K
21tok/sest.
Memory10.8 / 11.4 GB$1,104
What runs where
Every popular model against every machine, at 32k context. Each cell is the estimated speed in tokens per second.
- 30+ tok/s, fast enough for agents
- 15 to 30, fine for chat and coding
- Under 15, fits but slow
- Spills into system RAM
- Won't run
| Model | RTX 5060 Ti 16GB16 GB | RTX 5070 Ti16 GB | RTX 309024 GB | RTX 409024 GB | RTX 509032 GB | Mac mini M4 Pro 64GB64 GB | Mac Studio M4 Max 128GB128 GB |
|---|---|---|---|---|---|---|---|
| Gemma 4 26B-A4BUD-Q4_K_M · 4B active | 31offload | 44offload | 99 | 107 | 190 | 30 | 61 |
| Qwen 3.8 27BUD-Q4_K_M | 9offload | 11offload | 31 | 33 | 59 | 9 | 19 |
| Gemma 4 31BQ4_K_M | 6offload | 7offload | 27 | 29 | 51 | 8 | 16 |
| GLM 4.5 AirQ4_K_M · 13.5B active | won't run | won't run | won't run | won't run | won't run | won't run | 14 |
| gpt-oss 120BMXFP4 · 5.8B active | won't run | won't run | won't run | won't run | won't run | won't run | 49 |
| Mistral Small 4 119BUD-Q4_K_M · 7.1B active | won't run | won't run | won't run | won't run | won't run | won't run | 42 |
| Nemotron 3 Super 120B-A12BUD-Q4_K_M · 12B active | won't run | won't run | won't run | won't run | won't run | won't run | 25 |
| Qwen 3.5 122B-A10BQ4_K_M · 10B active | won't run | won't run | won't run | won't run | won't run | won't run | 30 |
Estimates from memory bandwidth and model size, not measurements. How estimates work
Popular GPUs
Compared on Qwen 3.8 27B at 32k context, the model people ask about most.
| GPU | VRAM | Bandwidth | Qwen 3.8 27B @32k | Price | Actions |
|---|---|---|---|---|---|
NVIDIA GeForce RTX 3090 nvidia | 24 GB | 936 GB/s | 31tok/sest. | $1,600 | Add |
NVIDIA GeForce RTX 5090 nvidia | 32 GB | 1,792 GB/s | 59tok/sest. | $4,921 | Add |
NVIDIA GeForce RTX 4090 nvidia | 24 GB | 1,008 GB/s | 33tok/sest. | $2,073 | Add |
| 16 GB | 896 GB/s | offload11 tok/sest. | $995 | Add | |
| 16 GB | 448 GB/s | offload9 tok/sest. | $618 | Add | |
| 24 GB | 960 GB/s | 28tok/sest. | $1,250 | Add | |
| 12 GB | 360 GB/s | offload5 tok/sest. | $370 | Add | |
| 16 GB | 640 GB/s | offload9 tok/sest. | $810 | Add |
New this week
The catalogue updates every day as models, hardware and prices change.
New models
- GrGranite Switch 4.1 30BAdded 2026-10-26
- WeWeVisOne VL 2BAdded 2026-10-26
- MiMinistral 3 14B Reasoning 2512Added 2026-10-25
- MiMinistral 3 3B Instruct 2512Added 2026-10-25
- MuMuse Glimmer 30BAdded 2026-10-25
New hardware
- NVIDIA RTX 6000 Ada GenerationAdded 2026-10-26
- RTX 6000 Ada GenerationAdded 2026-10-26
- Crucial T500 2TBAdded 2026-10-19
Price drops
- GMKtec EVO-X2 Ryzen AI Max+ 395 128GBAmazon$1,600$3,50054% off
- NVIDIA GeForce RTX 3090 TiAmazon$995$1,84046% off
- AMD Radeon RX 9060 XT 16GBAmazon$470$55015% off
- Intel Arc A770 16GBAmazon$310$35011% off
- Gigabyte B850 Aorus Elite WiFi7eBay$160$18011% off
More hardware for local AI
Already own a GPU?
Pick your card, Mac or mini PC and see every model it runs, at every context length, before you spend anything on an upgrade.
- Graphics cards
- Macs
- Mini PCs