AI API Router Directory

DigitalOcean's AI stack is GPU Droplets plus a managed Inference product (model catalog, serverless and dedicated inference, agents and knowledge bases) rather than one hosted LLM endpoint. GPU Droplets bill per second from $0.76/GPU/hour (RTX 4000 Ada) up to $11.19/GPU/hour spot or $7.94 reserved for HGX B300, and serverless inference is prepaid only, charging published per-token rates such as Llama 4 Maverick at $0.20 input / $0.696 output per 1M tokens and gpt-oss-120b at $0.055 / $0.385. New accounts can receive a $5 credit valid for 90 days.

A $5 new-account credit applied to usage and expiring after 90 days is described in the 'How does the $5 free credit work?' FAQ on the GPU Droplets pricing page; no larger AI-specific credit is stated on the official pages fetched.

Databricks Foundation Model Serving prices inference in Databricks Units (DBUs) on a pay-as-you-go basis with per-second granularity, no up-front cost and no published monthly platform floor. Per-model DBU rates per 1M tokens are public: Llama 3.3 70B at 7.143 input / 21.429 output, GPT-OSS-20B at 1.000 / 4.286, Llama 3.1 8B at 2.143 / 6.429, and Kimi K3 at 42.857 / 214.286 with a 4.286 cache-read rate. Databricks states the DBU price on the Serverless Real-time Inference SKU is $0.07 per DBU in the base region (us-east-1 for AWS, us-central1 for GCP, eastus for Azure), and the page offers a 14-day free trial without disclosing a fixed credit amount.

No dollar-denominated free credit is published: the pricing page offers a 14-day free trial and states that trial credit eligibility, amount and expiry are determined solely by Databricks after account creation (checked 2026-09-16).

Poixe AI is a commercial AI API router and aggregator that provides a unified interface for multiple LLM upstreams. It features model-group specific rate limits and a dashboard for API key management, targeting developers seeking simplified access to global and Chinese models.

No public free-tier information found in the source reference; treat free-tier availability as unverified rather than confirmed absent.