Baseten and denv are both inference platforms. The GPU bill is the split.

Baseten is a production inference platform. It is closer to denv than a raw token API is. Both exist so a team can serve a model without building every control-plane piece from scratch. The useful comparison is who owns the GPUs and the cluster, not a slogan.

What denv will say about itself

• Your provider account or your servers. Lambda, Nebius, EC2, Hetzner, bare metal. You pay them.
• denv does not resell GPUs and does not sell per-token inference. The denv bill is the cluster and the GPUs under management. See pricing.
• It is real Kubernetes. kubectl and Helm work. The line we use internally is “like EKS on top of EC2, but for inference.”
• Health checks, driver reinstall, MIG and DRA, failover are part of the product, and they are written down because they are easy to hide in a demo.

What this page will not say about Baseten

It will not claim Baseten cannot run custom models, autoscale, or offer a dedicated GPU. Those are questions for Baseten’s current docs. It will not invent a price comparison. It will not claim a speed win. Baseten was not a row in the published H100 board.

If your requirement is “a vendor runs the GPUs and bills me for the endpoint,” Baseten-shaped products match that, and denv does not. If your requirement is “the GPUs stay in accounts we already have, and the endpoint has to span more than one of them,” start with multi-cloud.

denv is early. That belongs in the comparison, not in a footnote.