FriendliAI review

Input tokens are billed at $0.14 to $1.40 and output tokens at $0.40 to $4.40, with rates differing by model. The catalogue covers GLM 5.3, GLM 5.3 Flash and GLM 5.2, so cheaper small models and larger ones sit on the same account.

Dedicated GPU hours start at $2.90, above the €1.40 category median. You choose from A100 80GB, H100 80GB and H200 141GB, so the hardware tier follows the model you plan to serve.

Billing granularity is per second, and dedicated endpoints are sold as hourly plans. Short bursts stop accruing charges the moment the job ends, which suits spiky workloads with idle gaps between jobs.

Yes. The API is OpenAI-compatible, so existing OpenAI client code can point at FriendliAI without a rewrite. Model access covers GLM 5.3, GLM 5.3 Flash and GLM 5.2, and the largest context window is 1,000,000.

The certifications listed are SOC 2 Type II and HIPAA, covering security audits and health data handling. The contract sits with FriendliAI Corp., headquartered in the US, a company founded in 2021. Regulated buyers usually ask for both before signing.

On entry input price it ranks at 52, and on output price at 43. The overall score is 4.9, split into cost at 4.2 and offer at 5.6, both drawn from the site's own scoring.

Support runs on email, help centre and community. A help centre and a community forum cover everyday questions, and email reaches the team for anything needing a person. Live chat and phone are not listed.

It suits teams moving off a single first-party model, since GLM 5.3, GLM 5.3 Flash and GLM 5.2 covers open-weight and speech models behind a single OpenAI-compatible endpoint. It also suits buyers who want A100 80GB, H100 80GB and H200 141GB set aside for their own workloads.