Model hosting and inference
Each provider is profiled from its own pricing page, with the source recorded against every figure. One counts people and ignores devices, one counts both and bills every machine past an allowance, and one counts people and then caps how many you may have. A small team with a large fleet lands in a different order from a large team with a laptop each.
LLM APIs
Model hosting and inference
GPU compute
| Provider | Entry paid plan | Free allowance | What is metered |
|---|---|---|---|
| Baseten | $0 per month on the Basic plan — pay as you go, with dedicated deployments, model APIs and training included rather than gated behind a fee. | No free plan | No platform fee: per-token Model APIs for open models, and dedicated deployments billed per minute of compute with volume discounts. The Pro tier buys priority access to high-demand GPUs rather than a lower rate. |
| Modal | $0 a month on Starter — the plan carries no fee at all, and $30 of monthly compute is included before anything is billed. The first plan with a monthly fee is Team at $250. | No free plan | A platform fee plus metered compute, and the fee buys a compute allowance rather than access. Starter is $0 a month and includes $30 of compute; Team is $250 a month and includes $100. Compute is billed per second on top of both. |
| Replicate | Not documented | No free plan | Two meters, chosen by the model rather than by you. Public models bill per run — per output image, per thousand tokens, per second of video. Private models you deploy yourself run on dedicated hardware and bill for THE TIME THEY SPEND SETTING UP, THE TIME THEY SPEND IDLE and the time they spend active. Fast-booting fine-tunes are the exception and bill active time only. |
| RunPod | No plan tiers: a new account can start with as little as $10 in prepaid credits, and deploying a Pod requires at least one hour of credits for the chosen configuration. | No free plan | Prepaid credit balance drawn down by usage: all compute and storage billed per second, with no data transfer fees and no monthly plan tiers. |
Which to consider
Teams that want per-token APIs and dedicated GPUs from one vendor with no plan fee — minute-granularity billing means a model that runs for ten minutes costs ten minutes.
Bursty, self-deployed inference where the bill should follow the work. Compute meters per second and per CPU core — a physical core is $0.0000131 per core-second with a 0.125-core minimum — and the H100 hour is $3.95, the cheapest of the four here that publishes one.
Running someone else's open model without packaging anything. Thousands of public models bill per run — per output image, per thousand tokens, per second of video — so a workload that fires occasionally pays only for the firing, with no instance to keep warm.
Short experiments and bursty inference — per-second billing and a $0.27 entry card mean an idle hour costs nothing and a small job is genuinely small.