Claude Haiku 5.5 Release Notes

Claude Haiku 5.5: High-Speed, Cost-Efficient Intelligence

Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks and is the fastest and cheapest small model released by Anthropic to date. It is specifically optimized for repetitive workloads such as summarization, data compaction, database queries, and classification requests, as well as speed-sensitive applications like live customer support and browser-based automation.

Performance and Benchmarks

Haiku 5.5 demonstrates significant capability gains over its predecessor, Haiku 4.5, and competes strongly with GPT-6 Luna. According to Anthropic's internal evaluations, Haiku 5.5 outperforms Haiku 4.5 across all listed benchmarks, including:

Knowledge Work: Scored 1620 on GDPval-AA v2.1 and 1578 on AA-Briefcase v1.1.
Computer Use: Achieved 72.4% on the OSWorld 2.1 offline subset, compared to 15.7% for Haiku 4.5.
Reasoning: Scored 45.9% on Humanity’s Last Exam (no tools).
Agentic Coding: Achieved 39.2% on Terminal-Bench 4.0 and 46.4% on FrontierCode 1.1 (Main).
Visual Reasoning: Scored 46.4% on Chartography (no tools).

For the first time in the Haiku line, the model includes an adjustable effort setting, allowing users to trade off between cost and intelligence based on the specific requirements of the task.

Tiered Pricing Model

Anthropic has introduced a tiered pricing structure for Haiku 5.5 based on prompt length. For prompts up to 100,000 tokens, the model is approximately 90% cheaper than Haiku 4.5. For prompts exceeding 100,000 tokens, the cost increases.

Price per 1M tokens	Prompts ≤ 100k	Prompts > 100k	Haiku 4.5	Sonnet 5.5
Cache reads	$0.01	$0.05	$0.10	$0.10
Cache writes	$0.125	$0.625	$1.25	$2.50
Input tokens	$0.10	$0.50	$1.00	$2.00
Output tokens	$0.50	$2.50	$5.00	$10.00

Ecosystem Updates: Sonnet 5.5 and API Credits

Beyond the Haiku 5.5 launch, Anthropic has implemented two major value improvements for the Claude platform:

Sonnet 5.5 Price Reduction: The cost of cache reads for Claude Sonnet 5.5 has been halved to $0.10 per million tokens, reducing the overall cost of most agentic tasks by approximately 20%.
Monthly API Credits: Max and Team subscribers now receive monthly API credits to facilitate the development of agents and applications. Max 5x users receive $100/month, Max 20x users receive $200/month, and Team subscribers receive up to $500 pooled across users.

Safety and Alignment

Haiku 5.5 features improved alignment over Haiku 4.5, with fewer instances of misaligned behavior and a lower willingness to cooperate with misuse. Cybersecurity safeguards are more restrictive than Haiku 4.5 but more permissive for defensive tasks than Sonnet 5.5. Biology safeguards remain consistent with the Sonnet 5, Sonnet 5.5, and Opus 5 models.

Community Insights and Technical Feedback

Early user feedback highlights a mix of performance gains and specific limitations:

Speed and Efficiency: Users report significant latency reductions. One engineer from Asana noted a 30% reduction in task completion latency and up to 2.5x faster inference per agent turn. Other users on OpenRouter report throughput around 131 tokens per second.
Capability Gaps: While strong for its size, some users found it lacking for complex UI generation. One tester noted that Haiku 5.5 correctly identified its own limitations and delegated a complex HTML task to Opus 5.5.
Pricing Concerns: Some developers have criticized the 100k token threshold as being too low for complex agentic workflows, noting that this boundary is quickly exceeded when managing long-term state or large contexts.
Tokenization Efficiency: Community members pointed out that the Haiku tokenizer may be less efficient than competitors like GPT-6 Luna, potentially consuming 1.5x more tokens for the same text, which impacts the effective cost.

"Haiku 5.5 beats Luna on every shared benchmark Anthropic published, particularly computer use and agentic coding."

"Ran our DataAnalyticsBench benchmark on it... 9x cheaper than Haiku 4.5 and 2 letter grades better... it's not as persistent as other models in its analysis."

Availability

Claude Haiku 5.5 is available now via the Claude Platform (claude-haiku-5-5), Amazon Web Services, Google Cloud, and Microsoft Azure. SDKs for Python and TypeScript have been updated to support computer use and browser use in beta.