Claude Haiku 5.5: Anthropic's cheapest, fastest small model, its price, benchmarks and what else changed

Claude Haiku 5.5 is the small model in Anthropic’s Claude 5.5 family, and it arrived on October 7, 2026, a little over a week after Claude Sonnet 5.5. Anthropic calls it “the cheapest, fastest, and most capable small model we’ve ever released”, and cuts the price sharply. This report covers the price, the benchmarks, the safeguards and the other changes announced alongside it, using Anthropic’s launch post and release notes. All benchmark figures are Anthropic’s own.

Screenshot: Anthropic’s Claude Haiku 5.5 announcement page. Captured by Universal Digital Assistant on October 17, 2026.

Claude Haiku 5.5 at a glance

What it costs

Anthropic’s pricing table compares Haiku 5.5 with the model it replaces and with Sonnet 5.5:

Anthropic says Haiku 5.5 is especially good value for prompts up to 100,000 tokens, which made up around 90% of requests to its previous Haiku. On average it costs about 75% less to run than Haiku 4.5.

What Anthropic says it is for

Haiku 5.5 “reliably handles quick and repetitive workloads”, such as summaries, compactions, database queries and classification, and pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work. As its fastest model, it also suits live customer support and browser use. Early-tester quotes in the post include:

• Asana: over a 30% reduction in latency and up to 2.5x faster inference per agent turn compared with the model it uses today.
• HubSpot: the best score it had seen from a smaller model on its CRM test suite, at 92.8%.
• Box: 11 points higher than Haiku 4.5 at about half the latency.
• Cognition: with Haiku 5.5 as a sidekick, its Devin Fusion holds a FrontierCode score of 66.2 while cutting cost and latency.

Benchmarks

The jump over Haiku 4.5 is large, and Haiku 5.5 beats OpenAI’s GPT-6 Luna on every test in this table that reports both. Anthropic is blunt about the limits, though: Sonnet 5.5 and Opus 5.5 “remain better choices for complex agentic coding tasks”, while Haiku 5.5 is best for narrowly scoped work that might otherwise have been too expensive.

Effort levels

Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, so developers can choose between cost and intelligence. Anthropic’s charts show its OSWorld and Terminal-Bench results at Low, Medium, High, Xhigh and Max effort.

Safety and safeguards

Anthropic says Haiku 5.5 shows major improvements across almost all alignment evaluations relative to Haiku 4.5, with far fewer instances of misaligned behaviour and lower willingness to cooperate with misuse. Its cybersecurity safeguards are more restrictive than Haiku 4.5’s but somewhat less restrictive than those on other recent models. They allow a wider range of defensive tasks than Sonnet 5.5’s safeguards but still block penetration testing and other attacker techniques. Its biology safeguards match Sonnet 5, Sonnet 5.5 and Opus 5, and organisations can apply to the Life Sciences and Cyber Verification Programs.

Other changes announced with it

• Sonnet 5.5 is cheaper. Cache reads on Sonnet 5.5 now cost 50% less, $0.10 instead of $0.20 per million tokens, which Anthropic says lowers its cost on most agentic tasks by about 20%.
• Monthly API credits for Max and Team. Claude’s release notes say Max and Team plans now include a monthly credit for running your own apps and agents on the Claude Platform, rolling out over a few days. You claim it by linking a Claude Console organisation in billing settings.

Where it fits

Claude’s lineup is now Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5; see Claude models explained for which to use, and Claude Pro vs Max for the plans. For coding workflows, see how to use Claude Code, and for the assistant overall, the Claude profile. Rival small models include OpenAI’s GPT-6 Luna (see GPT-6 Sol and Luna) and Google’s Gemini 4 Argon at the top end.