What is FriendliAI
FriendliAI is an inference platform for open-weight models with pay-per-token Model APIs, dedicated GPU endpoints and container deployment for private environments.
Model API pricing
As of September 25, 2026, the complete public Model API table lists GLM-5.3 at a temporary 10%-off rate of US$1.26 input, US$0.234 cached input and US$3.96 output per million tokens, with regular rates of US$1.40, US$0.26 and US$4.40. GLM-5.3-Flash costs US$0.15 input, US$0.03 cached input and US$0.50 output. GLM-5.2 and GLM-5.1 each cost US$1.40 input, US$0.26 cached input and US$4.40 output.
Gemma 4 31B costs US$0.14 input and US$0.40 output; DeepSeek V3.2 costs US$0.50 input, US$0.25 cached input and US$1.50 output; MiniMax M2.5 costs US$0.30 input, US$0.06 cached input and US$1.20 output. Whisper Large v3 speech-to-text is US$0.0015 per audio minute.
Dedicated and enterprise pricing
On-demand dedicated endpoints are billed per second with no startup charge. Through September 30 the hourly rates are A100 80GB US$2.90, H100 US$3.90, H200 US$4.50, B200 US$8.90 and B300 US$12. From October 1 the published rates become US$4, US$5, US$7, US$9 and US$12 respectively. BYOG and Enterprise are contact-sales routes.
Enterprise is a configurable contract rather than a fixed bundle. Published options include custom API rate limits, reserved GPU capacity, priority hardware access, custom regions, VPC or on-premises deployment, dedicated support, named customer success and custom commercial terms.
Official sources: FriendliAI pricing, FriendliAI documentation.