Baseten vs Koyeb

Baseten

Team API keys scoped to inference-only, metrics-only or a single environment or model, plus a Viewer role since 1 September 2026. H100 at $6.50 and A100 at $4.00 an hour, and start-up and idle replica time are billed.

Koyeb

H100 at $2.50 and H200 at $3.00 an hour, billed per second. Changelog silent since 27 February 2026 and no CLI release since 12 May 2026.

Baseten

1. Create a team key with inference-only permission for calling models and keep full-access keys out of the agent
2. Sleep for retry_after seconds on a 429 from api.baseten.co; the activate and deactivate endpoints allow 20 calls a minute
3. Retry 429, 503 and 529 with backoff, but treat 500 as a bug in your model code
4. Set scale_down_delay below the 900-second default or every burst bills 15 idle minutes
5. Send payloads over 256 KiB to /predict, not /async_predict, unless support has raised the async limit

Koyeb

1. Pass dry_run on service create or update to validate the definition before anything deploys
2. Page service lists with limit and offset and filter by statuses instead of fetching everything
3. Expect the first request after deep sleep to take 1 to 5 seconds and retry once with a timeout
4. Set --min-scale 0 on GPU services so idle instances stop billing
5. Re-check the Mistral transition before building anything long-lived on it