Anthropic released Claude Haiku 5.5 on October 7, 2026. The model targets high-volume, cost-sensitive work such as summaries, compactions, database queries, and classification. It also supports speed-sensitive uses such as live customer support and browser use. Anthropic positions it as a subagent for coding alongside Opus 5.5 and Sonnet 5.5.

At each model’s standard speed, Anthropic describes Haiku 5.5 as its fastest model to date. Opus models run faster in Fast Mode. Haiku 5.5 costs about 75% less to run on average than Haiku 4.5. The listed price is 90% lower for prompts up to 100,000 tokens and 50% lower above that limit. An updated tokenizer means Haiku 5.5 uses slightly more tokens per task.

Anthropic’s published benchmark figures use the order Haiku 5.5, Haiku 4.5, GPT-6 Luna, and Sonnet 5.5. GDPval-AA v2.1 scores are 1620, 735, 1437, and 1840. AA-Briefcase v1.1 scores are 1578, 614, 1336, and 1824. OSWorld 2.1 on its offline subset records 72.4%, 15.7%, 48.9%, and 83.9%.

Humanity’s Last Exam scores without tools are 45.9%, 10.2%, no result, and 56.9%. With tools, the figures are 57.4%, 18.7%, no result, and 64.5%. Terminal-Bench 4.0 scores are 39.2%, 0.0%, 16.4%, and 70.6%. FrontierCode 1.1 Main scores are 46.4%, no result, 42.4%, and 52.1% at Xhigh. Chartography scores without tools are 46.4%, 6.4%, 29.1%, and 61.6%.

OSWorld 2.1 measures computer agents on long, multistep tasks. GDPval-AA v2.1 evaluates professional work across 44 occupations. Humanity’s Last Exam tests expert academic knowledge and reasoning. Terminal-Bench 4.0 measures complex command-line tasks. Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, which lets users trade cost against capability. Anthropic shows the setting on OSWorld, GDPval-AA v2.1, and Humanity’s Last Exam.

Early customer tests reported lower latency and higher task scores. Asana’s Aaron Vinh reported more than 30% lower completion latency and up to 2.5 times faster inference per agent turn in AI Teammates evaluations. HubSpot’s Ze’ev Klapow reported a 92.8% average score across three CRM-task runs, plus the fastest completion, highest hit rate, and lowest false-positive rate in the tested set.

AlphaSense runs about 8 million Ask in Document calls per week. Daniel Campos said 400 test queries improved from 0.76 on Haiku 4.5 to 0.84 on Haiku 5.5, with a statistically significant difference. Box’s Yashodha Bhavnani reported an 11-point gain over Haiku 4.5 at about half the latency. Rogo’s Alex Wang described the model as suitable for quick lookups, summaries, and subagents. Cognition’s Walden Yan reported a 66.2 FrontierCode score for Devin Fusion with Haiku 5.5 as the sidekick, alongside lower cost and latency.

Prices are listed per 1 million tokens. Haiku 5.5 costs $0.01 and $0.05 for cache reads, $0.125 and $0.625 for cache writes, $0.10 and $0.50 for input, and $0.50 and $2.50 for output on prompts up to and above 100,000 tokens. Haiku 4.5 costs $0.10 for cache reads, $1.25 for cache writes, $1.00 for input, and $5.00 for output. Sonnet 5.5 costs $0.10, $2.50, $2.00, and $10.00 for the same categories. Around 90% of requests to the previous Haiku model were within the 100,000-token threshold.

Haiku 5.5 shows fewer misaligned behaviors and less willingness to cooperate with misuse across almost all of Anthropic’s alignment evaluations. Its cybersecurity safeguards are stricter than Haiku 4.5’s and less restrictive than those on several recent models. They allow more defensive work than Sonnet 5.5’s safeguards, but block penetration testing and techniques more likely to help attackers. Biology safeguards match those for Sonnet 5, Sonnet 5.5, and Opus 5. They allow research questions and restrict requests judged likely to cause harm. Anthropic directs organizations with broader biology or cyber requirements to its Life Sciences Verification Program and Cyber Verification Program.

Claude Haiku 5.5 is available on Anthropic’s platform, Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can use the model ID claude-haiku-5-5. Anthropic provides a migration guide.

Anthropic also cut Sonnet 5.5 cache-read pricing from $0.20 to $0.10 per million tokens. The company expects the change to reduce Sonnet 5.5 costs by about 20% on most agentic work. Max 5x subscribers will receive $100 in monthly API credits, Max 20x subscribers $200, and Team subscribers up to $500 pooled across users. The credits are rolling out during the week of the launch and work with any Claude model on the Claude Platform.

Anthropic is adding beta computer-use and browser-use support to its Claude Python and TypeScript SDKs. The company highlights Haiku 5.5 for these tasks because of its speed, capability, and price. Sonnet 5.5 and Opus 5.5 remain the stronger choices for complex agentic coding measured by Terminal-Bench 4.0.