Anthropic cuts Haiku 5.5 pricing by up to 90% as smallest model gains agentic capabilities

Anthropic unveiled Claude Haiku 5.5 on Wednesday, marking the first update to its smallest and least expensive model in nearly a year. The release represents Anthropic's third 5.5-generation model launch within a month, though Fable 5.5 remains absent from the lineup—likely undergoing extended review processes before becoming available.

Pricing structure and cost reductions

Haiku 5.5 delivers substantially lower pricing compared to its predecessor. Requests under 100,000 tokens now cost $0.10 per million input tokens and $0.50 per million output tokens, down from Haiku 4.5's flat $1/$5 rate. Larger requests are priced at $0.50/$2.50, compared to Haiku 4.5's unchanged pricing regardless of request size. Anthropic notes that approximately 90% of Haiku 4.5 requests fell into the lower-priced tier, suggesting typical developer usage patterns favor smaller batches. These adjustments represent 90% and 50% per-token reductions respectively, with Anthropic citing an average savings of around 75% when accounting for a mix of request sizes and an updated tokenizer that consumes slightly more tokens per task.

Haiku 5.5 introduces effort controls for the first time, with medium as the default setting, enabling developers to regulate token consumption on specific tasks. "Anthropic describes Haiku 5.5 as its 'fastest and most efficient model,' but this time, it is also much cheaper."

Use cases and workload targets

Traditionally, compact models have handled high-volume operations including summarization, classification, and routing. Anthropic positions Haiku 5.5 as capable of handling emerging workloads such as data compaction, database queries, and agent-based tasks where latency matters—including live customer support interactions and browser automation.

Performance benchmarks

Haiku 5.5 demonstrates substantial improvements over Haiku 4.5 across multiple evaluation frameworks. On the offline subset of OSWorld 2.1, which measures computer-use capabilities, Haiku 5.5 achieved 72.4% compared to Haiku 4.5's 15.7%, surpassing OpenAI's GPT-6 Luna at 48.9%.

On the GDPval-AA v2.1 knowledge-work benchmark, Haiku 5.5 scored 1,620 versus Haiku 4.5's 735 and GPT-6 Luna's 1,437. Terminal-Bench 4.0, which evaluates complex multi-step command-line operations, shows Haiku 5.5 at 39.2%, compared with 0% for Haiku 4.5, 16.4% for GPT-6 Luna, and 70.6% for Sonnet 5.5. Sonnet 5.5 maintains the lead across both benchmarks.

Competitive landscape with Chinese models

Anthropic's published comparisons focus on its own models and OpenAI's GPT-6 Luna, but developers evaluating small models also consider alternatives from Z.ai, Alibaba, and other vendors. Artificial Analysis rates Z.ai's GLM-5.3-Flash at 1,647 on GDPval-AA v2.1 and 1,454 on AA-Briefcase v1.1, compared to Haiku 5.5's reported 1,620 and 1,578 respectively.

Lower-cost options exist in the market. Alibaba's international offering of Qwen3.7 Flash costs $0.03 per million input tokens and $0.13 per million output tokens for inputs up to 32,000 tokens. For inputs between 32,000 and 256,000 tokens, pricing rises to $0.10 and $0.40 respectively.

Availability and deployment

Haiku 5.5 is accessible through the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. On Anthropic's platform, developers reference the model as claude-haiku-5-5. Beta support for computer use and browser use is rolling out to Anthropic's Python and TypeScript SDKs.

Additional pricing and subscription updates

Alongside Haiku 5.5's launch, Anthropic halved Sonnet 5.5's cache-read pricing from $0.20 to $0.10 per million tokens. "Anthropic is cutting Sonnet 5.5's cache-read price in half" to reduce agentic task costs by approximately 20%. This reduction is rolling out across platforms on Wednesday, though some existing Azure and Google Cloud customers will experience a delay of several days.

Anthropic is introducing monthly API credits to Max and Team subscription tiers this week. Max 5x subscribers receive $100 monthly, while Max 20x subscribers get $200. Team subscribers receive up to $500 pooled across their organization. Credits apply to any model on the Claude Platform.

Security considerations

Haiku 5.5 incorporates enhanced cybersecurity safeguards relative to Haiku 4.5, permitting a broader scope of defensive work than Sonnet 5.5's restrictions. However, penetration testing remains blocked. Organizations requiring expanded access for cybersecurity or biology applications can petition Anthropic's verification programs.