Anthropic releases Claude Haiku 5.5 with two-tier pricing by prompt length

A smaller model used to carry one price per million tokens for every call a team ran through it. Anthropic released Claude Haiku 5.5 on 7 October 2026, the first update to the Haiku tier in about a year, and set two separate prices on the Claude Platform: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 input and $2.50 output per million for prompts over that line.

Anthropic calls Haiku 5.5 the fastest and most efficient model in the Claude 5.5 family. The model is available on Claude.ai for Free, Pro, Max, Team and Enterprise users, on the Claude Platform, on Amazon Web Services, Google Cloud and Microsoft Foundry, and in Claude Code.

Effort controls arrive on the small tier

Haiku 5.5 is the first Haiku with Anthropic's effort controls, which let a team tune cost against intelligence for each task. Anthropic positions the model as a subagent paired with Opus 5.5 or Fable: the larger model plans, Haiku runs high-volume subtasks like summaries, classification, routing, and compaction. Anthropic also lists browser and desktop automation and small, specific code edits across many files.

Prompt caching saves up to 90 percent and batch processing saves 50 percent, both unchanged in Anthropic's pricing text.

Vendor-reported benchmarks and customer figures

Every number on Anthropic's page is Anthropic's own or a customer's. Anthropic says Haiku 5.5 is better than Haiku 4.5 across coding, tool use, computer use and agents, without publishing a benchmark table on the product page itself.

Named customer quotes on the page: Daniel Campos at an unnamed Ask in Document deployment running 8M calls a week reports a 0.84 against 0.76 score for Haiku 5.5 against Haiku 4.5 across 400 queries. Box AI reports Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency in early testing. HubSpot reports 92.8 percent averaged over three runs on a CRM evaluation suite. Cognition reports a 66.2 FrontierCode score for Devin Fusion with Haiku 5.5 as the sidekick under Opus 5.5. These are vendor-attributed figures and not independent evaluations.

The design question the two-tier price forces

The 100,000-token line changes what Haiku 5.5 costs to run as a subagent. Under the line, input and output cost $0.10 and $0.50 per million; over the line, $0.50 and $2.50. A team that keeps a subagent's prompt short, by routing the long context through retrieval or by giving Haiku only a small part of a conversation at a time, pays five times less per call than a team that hands Haiku a long context to summarise. The choice of a smaller model used to carry one price; it now carries two, and the design of the context decides which one.

Source

Anthropic, Claude Haiku page. Pricing and availability figures: Anthropic's own text.