Best Inference Providers

How we researched and ranked inference providers

Last updated: 2026-11-09

What we did

We built a shortlist of companies that sell API access to AI models (inference) in 2026, from current public sources. For each company we read its official pricing page, documentation, model pages and recent announcements on 2026-11-09. Some pricing pages build their tables with JavaScript, so we read them in a browser-rendered view or used the provider's own Markdown documentation and per-model pages instead. Every price and figure on this site traces back to one of those official pages, and we link to them.

What we did not do

Ranking criteria

We ranked providers on the following, roughly in order of weight:

  1. Breadth of current models: how many popular open-weight models are available self-serve today.
  2. Range of deployment options: serverless, reserved throughput, dedicated GPUs and custom weights, so a team can grow without switching vendors.
  3. Published price transparency and level: clear per-token or per-GPU rates, and how they compare on the same model (for example, gpt-oss-120b).
  4. Vendor-reported speed: considered, but weighted below breadth because these figures are not independent.
  5. Production and enterprise readiness: SLAs, data-retention policies, region controls and compliance certifications the provider documents.
  6. Momentum and stability: recent launches, funding and changes to the catalog or the company, as disclosed in official announcements.

We list gateways (OpenRouter and Hugging Face Inference Providers) last. They route to other providers rather than running inference themselves, so they solve a different problem.

Known limits

Updates and corrections

We review the list when providers change pricing, catalogs or products, and change the Last updated date when we do. If you spot an error, contact us and we will correct it.

Back to the ranked list · Pricing comparison · About