Beam vs Cerebrium

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.

Cerebrium

Per-second GPU prices are public, a 94-operation OpenAPI spec covers the management API, and service account tokens expire and are limited to named projects. Deployed endpoints are callable without a token unless disable_auth = false is set, and no request rate limits, 429 handling or SLA were found in the reviewed documentation.

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

Cerebrium

1. Set disable_auth = false in cerebrium.toml before deploying. The default leaves the endpoint callable by anyone with the URL
2. Authenticate headless with CEREBRIUM_SERVICE_ACCOUNT_TOKEN. cerebrium login opens a browser
3. Raise response_grace_period for long work. It defaults to 15 minutes and async runs stop at 12 hours
4. Send ?async=true to get a run_id with HTTP 202, and add webhookEndpoint because async calls return no result to the caller
5. Check the plan before choosing hardware. A100, H100, H200, B200 and RTX PRO 6000 need Standard, and protected compute bills at twice the listed rate