Anthropic has introduced Claude Haiku 5.5, a new small AI model aimed at high-volume work such as summarization, classification, database queries, document extraction, and customer support. The company says it is its fastest model to date and the first model in the Haiku family to offer adjustable effort controls.
The new setting lets developers choose how much reasoning effort the model should use for a task. Lower settings can prioritize speed and operating costs, while higher settings can devote more computation to difficult requests. Anthropic positions the feature as useful for teams that need to balance response quality, latency, and cost across different workflows.
From repetitive tasks to AI agents
Haiku models have traditionally handled short, repeatable jobs at scale. With Haiku 5.5, Anthropic is also targeting more capable agentic workflows, where an AI system takes several steps to complete a task. The company suggests using the model as a subagent alongside its larger Sonnet 5.5 and Opus 5.5 models, for example to search documents, retrieve facts, compact context, or complete narrowly defined coding tasks.
Anthropic reports substantial gains over Haiku 4.5 in its internal and external benchmark results. On the OSWorld computer-use benchmark’s offline subset, Haiku 5.5 scored 72.4%, compared with 15.7% for its predecessor. OSWorld tests whether AI agents can operate a computer to complete multi-step tasks.
On GDPval-AA v2.1, a benchmark covering professional work across 44 occupations, Haiku 5.5 received a score of 1,620. That was above the reported 735 score for Haiku 4.5 and 1,437 for OpenAI’s GPT-6 Luna, though below Sonnet 5.5’s 1,840. Anthropic also reports a 39.2% score on Terminal-Bench 4.0, which measures complex command-line tasks. Sonnet 5.5 remains notably ahead at 70.6%.
The results suggest that Anthropic sees Haiku 5.5 as a model for clearly scoped, frequent jobs rather than a replacement for larger systems on complex coding or long-running autonomous work.
Lower-cost access and new developer tools
Anthropic has changed Haiku’s pricing structure for requests up to 100,000 tokens, which it says represented about 90% of requests to Haiku 4.5. For those requests, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Requests above 100,000 tokens cost $0.50 for input and $2.50 for output per million tokens.
The company says Haiku 5.5 costs about 75% less on average than Haiku 4.5 after accounting for usage patterns and a revised tokenizer. It also cut the cache-read price for Sonnet 5.5 in half. Anthropic says that change reduces the cost of many agentic workflows using Sonnet by around 20%.
Haiku 5.5 is available through Anthropic’s Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can access it under the model name claude-haiku-5-5. Anthropic is also adding beta support for computer use and browser use to its Python and TypeScript software development kits.
Anthropic says the model has stronger cybersecurity safeguards than Haiku 4.5. The restrictions still permit a range of defensive security work, according to the company, but block penetration testing and other activities it considers more likely to be misused.
Sources
- Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released – Anthropic
- Anthropic prices Haiku 5.5 at $0.10/1M input and $0.50/1M output tokens for requests up to 100K tokens, and $0.50 and $2.50 above, vs. $1 and $5 for Haiku 4.5 – The New Stack
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
