Beam vs Hugging Face Inference Endpoints

Beam

Per-millisecond billing with cold starts and image pulls free, H100 PCIe at $3.50 and RTX 4090 at $0.69 an hour. No published request rate limits, 429 handling or SLA.

Hugging Face Inference Endpoints

OAuth scopes separate reading endpoints from writing them, both OpenAPI documents are public, and the unauthenticated /v2/provider route lists every instance with its hourly price. An account needs a payment method and credits before the first deployment, no rate limits or SLA were found for the management API, and the docs price table disagrees with the live list in places.

Beam

1. Check the response body for ok: false on gateway calls; a failure can arrive as HTTP 200
2. Don't pipe beam deploy --format json output into CI logs, since it contains the workspace token
3. Route anything over 180 seconds to a task queue and poll the task instead of holding the endpoint request
4. Set keep_warm_seconds deliberately; the 180-second endpoint default bills three minutes of GPU after every call
5. Pass gpu=["RTX4090", "A10G"] so a job still schedules when one type is out

Hugging Face Inference Endpoints

1. Call GET https://api.endpoints.huggingface.cloud/v2/provider first and pick an instance whose status is available. The docs table lists types the API marks deprecated or not available
2. Send X-Scale-Up-Timeout: 600 on requests to an endpoint that scales to zero, or handle 503 while the first replica starts
3. Set scaleToZeroTimeout yourself. The docs give a default of 1 hour and the OpenAPI document says 15 minutes
4. Pause or delete an endpoint when the job is done. Billing covers every minute a replica is initialising or running
5. Give the agent a fine-grained token or the read-endpoints scope unless it must deploy. Endpoints are private by default and take the same Hugging Face token as a bearer