The Claude API gives an application direct access to Anthropic's models through the Messages endpoint, billed per token with no subscription. Pick it when you're building software that needs Claude's output, not chatting yourself; use claude.ai or Claude Code for interactive work instead.
Other good fits
A team that wants a direct provider relationship, with its own rate limits and spend controls, rather than routing through a third party.
+ Pros
Read pros
The Claude API is Anthropic's Messages endpoint: send a model name, a token budget and a list of messages, and get back a structured response with text, a stop reason and usage counts. It's the same models that power claude.ai and Claude Code, but called directly from your own code with full control over the request.
Pick it when…
You're building a feature, not chatting: a summariser, a drafting assistant, a classifier, something that calls the model as part of a program.
You want a direct account with Anthropic, with your own spend limits and rate limits, instead of a third-party layer in between.
You need a specific model's exact behaviour (a pinned model ID) rather than whatever a wrapper app happens to expose.
− Cons
Read cons
A fluent, confident response can still get a detail wrong; check anything the application will act on, especially numbers and names. Model access and available rate limits depend on your account tier, and a documented spend-limit error isn't the same as a service outage, so don't retry blindly into one.
Pick something else when…
You want to chat with Claude yourself, not build with it.Claude (claude.ai) is the right place for that, with a free plan.
You want one account and one key across several model providers.OpenRouter removes the need to integrate each provider's API separately.
Your organisation already runs on a specific cloud and wants Claude billed through it. Claude is available through Amazon Bedrock, Google Cloud and Microsoft Foundry with their own IDs and lifecycle dates.
Avoid if
You want one account that can call several providers' models; a router like OpenRouter removes the need to integrate each API separately.
Connection
Send a POST request to https://api.anthropic.com/v1/messages with your API key in the x-api-key header, an anthropic-version header, and a JSON body naming a model (for example claude-sonnet-5), a max_tokens budget and your messages. The response's content is in text blocks, not the choices array other providers use, so a client built for another API needs adjusting, not just a new base URL.
Batch requests run at 50% off; prompt-cache reads cost a fraction of the base input price (as little as 2.5% on Fable, 5% on Opus, 10% on the rest). There's no subscription: you pay only for tokens used, but spend limits and rate limits apply at the account level, so a request can fail on a spend cap even when it's small.
Data handling
Requests and responses are processed on Anthropic's API infrastructure; nothing stays local just because you called it from your own machine. Each model has its own retirement date (none sooner than October 2026 for the current lineup), so pin a specific model ID for anything that needs stable, unchanging behaviour over time.
First request
Documentation-based suggestion — not run here.
Draft a two-sentence acknowledgement of a made-up enquiry: a chess club asking about a room for twelve people, with instructions not to confirm availability or invent a price. Send it with claude-haiku-4-5, the fastest and cheapest current model, since this task doesn't need more.
Done when the response keeps the number of people correct, invents no availability or price, and you've read the usage field to see what the request actually cost.
Links & evidence
Documentation-based · not hands-on · Sep 23, 2026
Rests on Anthropic's API getting-started guide, its models overview and its pricing page, read on 2026-09-23. No API request was sent for this review.
Claude API getting started · Sep 23, 2026 — Messages request format, authentication headers, response structure.
Claude models overview · Sep 23, 2026 — Current model lineup (Fable 5.1, Opus 5.5, Sonnet 5, Haiku 4.5), pricing, context windows, retirement dates.
Claude API pricing · Sep 23, 2026 — Per-model token prices, batch discount, prompt-cache read pricing.
OpenRouter pricing · Sep 23, 2026 — Comparison alternative for calling multiple providers from one integration.
Further review notes & alternatives
Alternatives and trade-offs
Claude wins when you want to use Claude yourself through a chat interface, not build against it.
OpenRouter wins when you want to call several providers' models from one integration.
Claude Agent SDK wins when you want a full agent loop (tools, permissions, sessions) built in, not just a single request-response call.
Call Claude models from Python, TypeScript, or REST applications.
Choose it when you need direct control over prompts and structured outputs.
It suits scripted batch work because you supply the surrounding workflow.
