by Baidu

Baidu: ERNIE 4.5 21B A3B Thinking.

text open weights laptop+ 21B params 131K ctx
Cheapest input
$0.07/M
on OpenRouter
Cheapest output
$0.28/M
on OpenRouter
Smallest GPU
1× Nvidia RTX A4000
$0.026/hr
Run it yourself

Cheapest hardware per quantisation.

Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr refreshed hourly from each provider's API.

Quantisation Cheapest GPU config Total VRAM Live $/hr tokens/sec
FP16
FP16 — half precision (default)
64 GB Compare →
FP8
FP8 — 8-bit float (Hopper / Blackwell)
32 GB $0.023/hr Compare →
INT4
INT4 — 4-bit integer (~4× VRAM saving)
16 GB $0.026/hr Compare →
Just need an API?
Skip the GPU rental and call a hosted endpoint instead. See access providers on the Overview tab →