compute.info

Provenance

Sources

Source links appear directly in every comparison table. This register explains what qualifies as canonical and where the published dataset originates.

The unified AI directory retains separate API pricing and model-card sources. Weight-access labels link to explicit creator evidence with their own retrieval dates; an API price page alone does not establish whether weights are closed. Download links lead to model repositories, while exact source-document links remain attached to each record.

Each AI card exposes readable pricing, creator and weight-license evidence where available. Its source disclosure identifies the subjects and claims supported by each document, including separate raw API links. A later license review does not update the price observation date.

AI API pricing

The table stores the exact source URL used for each record. Provider pages can change without notice; retrieval, completed check and review dates are shown separately when known. A missing check/review date is not replaced with the site build time.

Virtual machine pricing

The catalog generally selects meaningful entry, balanced, compute, memory, storage, Arm, GPU, and accelerator representatives instead of publishing every near-duplicate size. When a provider's official API is ingested comprehensively, every API-listed SKU is preserved and its regional availability is shown explicitly; unavailable SKUs do not receive fabricated price rows. Prices retain their named region or published scope, native currency, storage, transfer, hourly-cap, and tax qualifications rather than being forced into a single synthetic unit.

Open models

AI evidence URLs use an explicitly reviewed HTTPS host allowlist and exclude credential-bearing URLs. Host acceptance is not proof of authorship or of a particular claim: creator ownership and model-specific scope still need review. A price page alone is not license evidence. Provider/CDN documents and model cards support only their actual statements; code and weight licenses, gates, hosted checkpoint identity and lineage are checked independently. Forums, search snippets and comparison sites are discovery signals, not canonical AI prices or licenses.

Model sizes, licenses, context windows, and modalities come from official creator repositories on Hugging Face. Each catalog row links directly to the model card used.

Hardware-fit methodology