Claude Haiku 5.5 launches with a 1M context window and lower API rates

Anthropic designed Claude Haiku 5.5 for high-volume, cost-sensitive tasks. Its launch page names summaries, compactions, database queries, classification, subagents, live customer support and browser use as good fits. Anthropic also calls it the fastest model in its range at standard speed.

The direct Anthropic price is $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Above that threshold, the listed prices are $0.50 input and $2.50 output. Cache reads and writes have separate rates. Anthropic says the model is around 75% cheaper to run on average than Haiku 4.5, with the exact saving depending on prompt size and token use.

Claude Haiku 5.5 accepts text, images and files. AIMI records reasoning, tool use and structured-output support on the OpenRouter route. That makes it a practical candidate for small subagents, document triage and browser-oriented tasks where latency and cost matter more than frontier depth.

The route carries a one-million-token context window and a 128,000-token maximum output. OpenRouter also exposes adjustable reasoning effort from low through max. Those limits belong to the verified OpenRouter route, so check the active gateway before relying on them elsewhere.

Anthropic dated the official release page 7 October 2026. AIMI first saw the normal and batch OpenRouter routes plus OpenCode listings at 00:27:43 SAST on 8 October, 27 minutes and 43 seconds after the start of the SAST day. That was 46 minutes and 57 seconds before the first AZ Labs article package was created.

A fresh check during this update still lists Claude Haiku 5.5 on OpenCode Zen. The OpenCode Go endpoint no longer lists it, so that route remains a candidate rather than a current availability claim. These are gateway observations for one canonical model, not separate releases or evidence that a provider was responding to Anthropic.

Haiku 5.5 is a good first test for work that is frequent, bounded and easy to evaluate: summarising support threads, compacting context, classifying records, extracting fields and running narrow subagents. Anthropic’s own release page says Sonnet and Opus remain better choices for complex agentic coding tasks.

Provider behaviour still matters. Pricing, rate limits, retention, tool schemas and reasoning controls can differ between direct Anthropic access and OpenRouter. AZ Labs customers can use the AI Gateway as the integration layer, then compare the exact route against their own prompts before moving production traffic.