LLM APIs

LLM APIs sell access to large language models by the token: you send a prompt over HTTPS and pay for the input and output tokens each request uses. Product teams building chat, search, summarizing, coding and agent features use them, as do back-office teams automating document and support work. They differ in which models they sell and at what price, whether there is a free tier, what happens to your prompts, which certifications and data regions they offer, and how much batch processing, caching or off-peak hours take off the bill.

We compare them on the same six indicators, read from each vendor's pricing, model, data-usage, trust and documentation pages:

• Pricing clarity: whether per-token prices are published for every model and can be paid online, and whether the pages leave prices out or contradict themselves.
• Free tier: an ongoing free tier with published limits and no training on free-tier prompts, one-time free credits, or nothing.
• Price for 100M input + 20M output tokens: the listed monthly cost of 20,000 requests of 5,000 input and 1,000 output tokens on the vendor's lowest-priced generally available model with tool calling and a context of at least 128,000 tokens, with the arithmetic shown; 0 when the pages cannot price it.
• Data use and retention: whether paid API data is kept out of training by default, how long it is kept, zero data retention, and a data processing addendum with the terms.
• Compliance and regions: the vendor's own SOC 2 Type II and ISO 27001 for the API, a HIPAA BAA, and processing kept in a region you choose.
• Cost controls and developer tools: batch, caching and other lower rates, published rate limits, official SDKs, JSON-schema output, a compatible endpoint, spend limits and console SSO.

Model quality and speed are not scored: the pages are vendor statements, and we do not run the models. Each review says which pages it was written from and when. Where we have run a product hands-on, the evidence is on its page.