Claude Haiku 5.5 cuts small-model cost and becomes the default Haiku

Anthropic released Claude Haiku 5.5 on October 7, 2026, and called it the cheapest, fastest, and most capable small model it has ever released. On average it costs around 75% less to run than Haiku 4.5. The same day, Claude Code 2.1.293 made claude-haiku-5-5 the default Haiku model on the Anthropic API.

Haiku 5.5 is built for high-volume, cost-sensitive work: summaries, compaction, database queries, and classification. Anthropic says it pairs with Opus 5.5 and Sonnet 5.5 as a subagent on coding work, and that the speed suits live customer support and browser use. A footnote qualifies that speed claim. Haiku 5.5 is the fastest model at each model's standard speed, but it runs less quickly than Opus models in Fast Mode.

What the tokens cost

The list price splits at 100,000 tokens. Anthropic says prompts up to that length made up around 90% of requests to the previous Haiku.

Per million tokens, up to 100K / over 100K:

Input: $0.10 / $0.50, against $1.00 on Haiku 4.5
Output: $0.50 / $2.50, against $5.00
Cache reads: $0.01 / $0.05, against $0.10
Cache writes: $0.125 / $0.625, against $1.25

Haiku 5.5 is priced 90% lower than Haiku 4.5 for requests up to 100,000 tokens and 50% lower for requests over 100,000 tokens. The average of around 75% less to run also accounts for an updated tokenizer, similar to Sonnet 5.5's and Opus 5.5's, which uses slightly more tokens per task.

Claude Code 2.1.293 sets claude-haiku-5-5 as the default Haiku on the Anthropic API, lists a 1 million-token context, and cites $0.10/$0.50 per million tokens ($0.50/$2.50 for prompts over 100K). Those pairs match the input and output rows on the launch rate card.

Where it sits beside larger models

With the launch, Anthropic halved Sonnet 5.5 cache reads from $0.20 to $0.10 per million tokens. Cache reads are a large share of token consumption, so the company says the cut makes Sonnet 5.5 around 20% cheaper on most agentic tasks.

The same post draws the job split. Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding, including tasks like those in Terminal-Bench 4.0. Haiku 5.5 is for narrower work that used to be too expensive to run often: compaction, summarization, and subagent calls.

The published scores line up with that split. Terminal-Bench 4.0 is 39.2% for Haiku 5.5, 0.0% for Haiku 4.5, and 70.6% for Sonnet 5.5. OSWorld 2.1's offline subset is 72.4%, against 15.7% and 83.9%. GDPval-AA v2.1, Artificial Analysis's evaluation of professional work across 44 occupations, is 1620, against 735 and 1840. Anthropic points to the Haiku 5.5 system card for how those evaluations were run.

Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, so users can optimize for cost or intelligence. The launch charts plot Low, Med, High, Xhigh, and Max against cost on OSWorld, GDPval-AA, and Humanity's Last Exam.

Early customer notes on the same page track the speed pitch. Asana's Aaron Vinh, staff software engineer, said AI Teammates evals showed over a 30% reduction in task-completion latency and up to 2.5x faster inference per agent turn versus the model Asana uses today. HubSpot's Ze'ev Klapow, distinguished software engineer, said Haiku 5.5 scored 92.8% averaged over three runs on simulated CRM portals, the best score that suite had seen, and was the fastest model tested on a stale-record audit, with the highest hit rate and the lowest false-positive rate.

Safeguards and where to call it

Anthropic says Haiku 5.5 shows major improvements across almost all of its alignment evaluations relative to Haiku 4.5, with far fewer instances of misaligned behavior and a lower willingness to cooperate with misuse.

Cybersecurity safeguards are more restrictive than Haiku 4.5's and somewhat less restrictive than those on other recent models. They permit a wider range of defensive tasks than the Sonnet 5.5 safeguards, and they still block penetration testing and other techniques more likely to be used by attackers. Biology safeguards are the same as for Sonnet 5, Sonnet 5.5, and Opus 5. They allow research biology questions and restrict requests judged likely to cause harm. Organizations working on wider-ranging biology and cyber activities can apply to the Life Sciences Verification Program and the Cyber Verification Program.

Haiku 5.5 is available now on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. The launch page and Anthropic's @claudeai account both said so on October 7. On the Claude Platform the model id is claude-haiku-5-5. Anthropic points developers to a migration guide for the change.

If summaries, compaction, or subagents dominate your Anthropic API spend, read the Haiku 5.5 rate card and migration guide, then confirm Claude Code is on 2.1.293 before you set effort on prompts up to 100,000 tokens.

What the tokens cost

Where it sits beside larger models

Safeguards and where to call it

Sources