Claude Haiku 5.5

What it claims

Claude Haiku 5.5 (API ID claude-haiku-5-5) is Anthropic's new small model for high-volume, narrowly scoped work: summarization, context compaction, classification, database queries, and subagent duty under Opus 5.5 or Sonnet 5.5. Anthropic calls it its fastest model at standard speed, though it is slower than Opus models in Fast Mode. It is the first Haiku-class model with an adjustable effort setting. The headline "~75% cheaper than Haiku 4.5" is a blended figure:

• Under 100k tokens per request: the price is 90% lower than Haiku 4.5.
• Over 100k tokens per request: the price is 50% lower.
• The blend: it assumes Haiku 4.5's traffic mix, where 90% of requests were under 100k tokens. It also nets out a new tokenizer, similar to Sonnet 5.5's and Opus 5.5's, which uses slightly more tokens per task.

Evidence

The benchmark results and the per-effort charts are images, and their numbers are not in the page text. The only quantitative claims in the text are the pricing ratios above. Customer quotes are referenced but not reproduced.

How to use it

• Availability: the Claude Platform, AWS, Google Cloud and Microsoft Azure. Anthropic points to a migration guide.
• Where your savings land: if your workload has long prompts (over 100k tokens), expect roughly half-price, not 90% off. Re-measure token counts, because the tokenizer changed.
• Sonnet 5.5 cache reads: now $0.10 per million tokens, down from $0.20. Anthropic estimates this makes Sonnet 5.5 about 20% cheaper on most agentic tasks.
• Monthly API credits: usable on any model.
• Max 5x: $100
• Max 20x: $200
• Team: up to $500, pooled across users
• SDKs: the Python and TypeScript SDKs add computer use and browser use in beta. Haiku 5.5 is pitched as the default for these tasks because of its speed and price.

Limits

• Not for complex agentic coding. Anthropic states that Sonnet 5.5 and Opus 5.5 remain the better choice for tasks like those in Terminal-Bench 4.0. Haiku 5.5 is meant for tasks that were previously too expensive to run on Claude.
• Cybersecurity safeguards:
• Stricter than Haiku 4.5, but looser than other recent models.
• Allow a wider range of defensive tasks than Sonnet 5.5.
• Still block penetration testing and other techniques more likely to be used by attackers.
• Security teams planning offensive testing should note this restriction.
• Biology safeguards: the same as Sonnet 5, Sonnet 5.5 and Opus 5. Research questions are allowed, and requests judged likely to cause harm are restricted.
• Getting broader access: organizations can apply to the Life Sciences Verification Program or the Cyber Verification Program.
• Alignment: the page claims "major improvements" over Haiku 4.5 but gives no figures here. The details are in the system card.