Anthropic has released Claude Haiku 5.5, updating its fastest and most affordable small AI model to target high-volume operations such as document summarization, database queries, and customer support. The model is available immediately across Amazon Web Services, Google Cloud, and Microsoft Azure.
Lower Pricing and Tokenizer Adjustments
Anthropic states that Haiku 5.5 costs about 75 percent less on average than Haiku 4.5. For prompts up to 100,000 tokens—which Anthropic reports represents roughly 90 percent of previous Haiku queries—token prices drop by up to 90 percent, though requests exceeding 100,000 tokens cost five times as much. However, an updated tokenizer increases token usage per task compared to its predecessor, similar to the 30 percent token count bump observed in Opus 4.x updates. As a result, net operational cost savings will depend on specific prompt sizes and task compositions.
Benchmark Performance Gains
Haiku 5.5 achieves substantial score increases across multiple standardized evaluations compared to Haiku 4.5. On the knowledge benchmark GDPval-AA v2.1, it reaches 1,620 compared to the previous 735. On Humanity's Last Exam, its score improves from 10.2 percent to 45.9 percent without external tools, and from 18.7 percent to 57.4 percent with tools.
In automated computer navigation, Haiku 5.5 scored 72.4 percent on OSWorld-2.1, up from 15.7 percent. It also recorded 39.2 percent on the agentic coding evaluation Terminal-Bench 4.0, where Haiku 4.5 scored zero. In Anthropic's reported comparisons, Haiku 5.5 leads OpenAI's GPT-6 Luna budget model across all evaluated categories, though it remains behind Anthropic's Sonnet 5.5 reference benchmarks.
Adjustable Reasoning and Security Constraints
Haiku 5.5 is the first model in the Haiku tier to offer adjustable reasoning levels, enabling developers to tune output quality against token costs. Anthropic recommends the model for focused tasks like summarization, text compaction, and sub-agent workflows, while directing complex agentic coding to Sonnet 5.5 or Opus 5.5.
Cybersecurity controls on Haiku 5.5 are tighter than on Haiku 4.5, but permit a wider range of defensive tasks than Sonnet 5.5. Penetration testing capability remains blocked by default, with specialized verification programs available for life sciences and cybersecurity organizations requiring expanded access.
Sonnet Price Cuts and SDK Updates
Alongside the Haiku release, Anthropic halved cache read pricing for Sonnet 5.5 from $0.20 to $0.10 per million tokens, which the company estimates will lower total costs for typical agentic tasks by approximately 20 percent. The company is also offering recurring monthly API credits to subscribers—$100 for Max-5x, $200 for Max-20x, and up to $500 for Team accounts. Additionally, Anthropic updated its Python and TypeScript SDKs to add beta support for computer and browser automation.
AI-generated editorial illustration; not a documentary photograph.
Source: Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over
SIGNAL TITANS
Leave a Reply