vram.run Models Hardware Providers Cloud State of Inference
API provider data is live · Hardware & cloud pricing curated 2026-02-23

SmallThinker-3B-Preview

by Tiiny

3.4B params · text-generation · 415 likes · 188 downloads

SmallThinker-3B-Preview is a 3.4B parameter model. At Q4 quantization it requires 2GB of VRAM. It runs comfortably on GeForce RTX 4090 (386 tok/s), GeForce RTX 5090 (579 tok/s), M4 Max 128GB (141 tok/s).

Inference providers

Provider$/1M in$/1M outThroughput
Featherless

GPU compatibility

GPUVRAMQ4 DecodeVerdict
GeForce RTX 409024GB386 tok/scomfortable
GeForce RTX 509032GB579 tok/scomfortable
M4 Max 128GB128GB141 tok/scomfortable
M4 Pro 48GB48GB70 tok/scomfortable
M4 Pro 24GB24GB70 tok/scomfortable
A100 PCIe 80 GB80GB708 tok/scomfortable
H100 SXM5 80 GB80GB1424 tok/scomfortable
GeForce RTX 309024GB341 tok/scomfortable
Radeon RX 7900 XTX24GB282 tok/scomfortable
GeForce RTX 408016GB274 tok/scomfortable
Install CLI [email protected] Raw data · MIT · API data: live · HW/Cloud data: curated 2026-02-23 · v0.6.0