Claude 5 API release date: what's live right now

Quick answer: one Claude 5 model has shipped, not a full generation.

"Claude 5" is not one release. Anthropic ships each size in a generation on its own schedule, and right now only one of the three has a stable model ID and a published price: claude-sonnet-5. Opus and Haiku have not moved off the 4.x line. That makes a search for the Claude 5 API release date really three separate questions bundled into one. The Sonnet-size question already has an answer: it shipped, it has an ID you can call today, and it is billed at Anthropic's own list rate. The Opus-size and Haiku-size questions are still open, and there is no date attached to either one. The rest of this page treats those as separate tracks: what is confirmed today, what a call to Sonnet 5 actually costs, and how to set up an integration so an eventual Opus 5 or Haiku 5 release is a one-line change instead of a rewrite.

Why Anthropic ships tiers separately instead of all at once.

None of this is specific to Anthropic. Every lab that ships multiple sizes under one generation name stages the rollout, and the numbering rarely lands on the same day across sizes. The reliable signal is the catalog entry for the exact model ID you plan to call, not a generation-wide announcement that may only describe one size of it.

• Sonnet is the default workhorse size, so it typically gets the first stable release inside a new generation.
• Opus carries the heaviest evaluation and safety review of the three sizes, which routinely pushes its release weeks or months behind Sonnet.
• Haiku is usually distilled from the larger models after they stabilize, so it tends to ship last rather than first.
• Model catalogs and pricing pages update per model ID, not per generation, so "Claude 5 pricing" really means whichever Claude 5 model IDs exist on the day you check.

Worked example: what a Sonnet 5 request actually costs.

Sonnet 5 is billed like any other Claude model: a rate per million input tokens and a separate rate per million output tokens, with the total scaling directly with request size. The table below prices three representative request sizes for claude-sonnet-5 at Anthropic's own list rate and at a lower per-token rate available through third-party gateways.

Why the release-date question keeps resurfacing.

A generation number does not map to one launch event: Sonnet, Opus, and Haiku each ship on their own schedule under their own model ID, so a "Claude 5" mention can sit on a page for months before every size in that generation exists. Default model aliases sometimes start resolving to a new tier without any announcement, which reads like a release even when none went out. The catalog entry for the exact model ID you plan to call is the fact that changes your integration and your bill; a generation-wide announcement is marketing copy layered on top of it.

Where each Claude 5-era model stands right now.

"Claude 5" today means one shipped model sitting alongside four carryover models from the previous generation. The table below is the current status of all five, pulled from the same catalog that prices every model on this site.

How to prepare your integration for a staggered rollout.

Treat the model ID as the unit of change, not the generation name. That is the one habit that keeps a staggered, size-by-size rollout from turning into an emergency migration later.

• Pin the exact model ID, claude-sonnet-5, instead of a generation alias, so a future Opus 5 or Haiku 5 release cannot silently change your default model.
• Add a fallback path: if a call to claude-sonnet-5 errors for your account or region, retry against claude-sonnet-4-6 instead of failing the request outright.
• Budget off today's published rate, not a guess. Sonnet 5 already costs less per token than Sonnet 4.6 at official list price; there is no published number yet for Opus 5 or Haiku 5, so do not bake one into a forecast.
• Watch the context window field in the catalog, not just the price. All five models above currently list a 1M-token window, so a version bump alone will not force a chunking rewrite.
• If you are testing tiers side by side to control cost, route the cheaper fallback calls through an OpenAI-compatible gateway instead of paying official list rates on every retry.
• Re-check the catalog before a larger migration. A model ID that does not exist yet, such as an eventual opus-5 or haiku-5 slug, should fail cleanly rather than silently falling through to a different tier.

Config example: call Claude Sonnet 5 today.

Because Sonnet 5 is already live, pointing an existing client at it is a model-name change, not an SDK upgrade. APIsRouter mirrors the OpenAI chat completions shape at https://api.apisrouter.com/v1, so the same request structure already in use for claude-sonnet-4-6 works by swapping the model field. Verify the ID resolves before wiring it into a default:

curl https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer sk-APIsRouter-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "messages": [{"role": "user", "content": "Reply with one word: ok"}],
    "max_tokens": 16
  }'

FAQ