Claude Haiku 5.5 drops to $0.10 per million input tokens

Developers who use Anthropic's API will pay $0.10 per million input tokens for Claude Haiku 5.5, compared with $1.00 for Haiku 4.5, according to the company's price list. The discount applies to requests of up to 100 000 tokens; above that size, the cut is 50%. A token is the unit these APIs use to bill the text that goes in and comes out.

Anthropic introduced the model on Oct. 7, 2026. It is available on its Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure under the identifier `claude-haiku-5-5`. It is the first Haiku with an effort setting, an option to trade cost for reasoning capability.

The company recommends Haiku 5.5 for summarizing, compacting context and working as a subagent, and acknowledges that Sonnet 5.5 or Opus 5.5 are a better fit for complex agentic coding. It also said the new tokenizer uses somewhat more tokens per task, a caveat that can trim the real savings.

What has not been tested independently

Almost all the figures come from Anthropic. On the OSWorld 2.1 test, for example, the company reports 72.4% for Haiku 5.5 and 15.7% for Haiku 4.5. Only GDPval-AA, where Haiku 5.5 scores 1620 Elo points, is evaluated by Artificial Analysis, according to the launch page; this outlet did not check the result with that firm.

Help Net Security reports, based on the system card, that Haiku 5.5 resisted prompt injection, an attack that hides text meant to give orders to the model, less well than Sonnet 5.5 and Opus 5.5 in Gray Swan's test, mostly in graphical computer use. Anthropic ran those evaluations.

The page gives neither a context window nor a price in Mexican pesos, and no Mexico-specific availability was found. It remains to be seen whether the savings hold up with the new tokenizer on real workloads.

Disclosure: part of El Mediático's operation runs on Anthropic models.