Anthropic announced the release of Claude Haiku 5.5, positioning it as the company’s most affordable and speedy small-scale model to date. The update targets developers needing a cost-effective solution for repetitive, low-creativity workloads.

Pricing and Availability

The new model charges ten cents per million input tokens and fifty cents per million output tokens for prompts that stay under 100,000 tokens. This represents a dramatic reduction—about ninety percent—from the previous Haiku 4.5 rates of $1 and $5 respectively. Anthropic estimates an average cost saving of roughly seventy-five percent because most requests fall below the token limit. For larger inputs, the price is halved again. The service is instantly accessible through Anthropic’s website and major cloud providers, including AWS, Google Cloud and Microsoft Azure, under the identifier claude-haiku-5-5.

Performance Benchmarks

Independent tests show Haiku 5.5 delivering notable gains over OpenAI’s comparable small model, GPT-6 Luna. On the OSWorld 2.1 suite, which measures an AI’s ability to operate a computer via multi-step commands, Haiku 5.5 achieved a 72.4% success rate, surpassing Luna’s 48.9%. In the Terminal-Bench 4.0 evaluation—focused on completing professional tasks through typed commands—Haiku 5.5 recorded 39.2% correct first-attempt completions, well ahead of Luna’s 16.4% and the earlier Haiku 4.5, which scored zero. By contrast, Anthropic’s larger Claude Sonnet 5.5 posted a 70.6% score on the same coding benchmark.

Target Use Cases

Anthropic markets Haiku 5.5 for high-throughput operations such as summarising lengthy documents, handling customer-support chats, and performing quick database queries. The model also supports latency-sensitive scenarios like live assistance and autonomous web-browsing, where response speed is critical. While the model remains capable of logical reasoning, Anthropic cautions users to verify outputs, noting occasional inaccuracies in simple queries.

Additional Offerings

Alongside the price cut for Haiku 5.5, Anthropic reduced the cache-read fee for its larger Claude Sonnet 5.5 to ten cents per million tokens. The company also introduced monthly API credit bundles for its Max and Team subscription tiers, ranging from $100 to $500 in pooled credits, to further lower entry barriers for developers.

Why it matters

By slashing costs and improving speed, Haiku 5.5 lowers the economic threshold for integrating AI into routine business processes. The model’s competitive benchmark results suggest that smaller, cheaper AI can now rival larger, more expensive alternatives for many practical tasks. This shift could accelerate adoption across SaaS platforms, customer-service operations and data-analysis pipelines, while also intensifying pricing pressure on rival providers such as OpenAI.