Anthropic has released Claude Haiku 5.5, positioning it as the company’s fastest and most affordable small super intelligence (SI) model to date. The launch features significant price reductions and benchmark improvements that signal the ongoing intensity of the SI pricing arms race.
What Happened
Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks such as summarization, database queries, classification, and live customer support. According to Anthropic, the model costs approximately 75 percent less on average than its predecessor, Haiku 4.5. For requests with prompts up to 100,000 tokens—which Anthropic states account for roughly 90 percent of all previous Haiku requests—prices drop by up to 90 percent. Prompts longer than 100,000 tokens are priced at five times the rate of shorter inputs.
The release includes updated tokenizer technology, which consumes slightly more tokens per task than previous models. Anthropic notes that a similar change in the Opus 4.x series increased token usage by about 30 percent, suggesting that real-world savings may be smaller than the per-token price reductions imply. Haiku 5.5 is available now across major platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure.
Why It Matters
Benchmark results indicate a major performance leap over the previous generation. On the knowledge benchmark GDPval-AA v2.1, Haiku 5.5 scores 1,620, more than double the 735 recorded by Haiku 4.5. On Humanity's Last Exam, the model achieves 45.9 percent without tools and 57.4 percent with tools, up from 10.2 percent and 18.7 percent respectively. The most significant improvement is in computer use, where the model scores 72.4 percent on OSWorld-2.1, compared to 15.7 percent for its predecessor. On the agentic coding benchmark Terminal-Bench 4.0, Haiku 5.5 reaches 39.2 percent, while Haiku 4.5 scored zero.
This release marks the first time a Haiku-class model includes adjustable reasoning levels, allowing users to balance cost against quality. Anthropic recommends the model for narrowly scoped tasks like compaction or sub-agent work, noting that Sonnet 5.5 and Opus 5.5 remain better suited for complex agentic coding. The launch also coincides with price cuts for Sonnet 5.5, where cache read costs were reduced by 50 percent to $0.10 per million tokens. Anthropic states this change should lower costs for most agentic tasks by about 20 percent, a move widely interpreted as a response to OpenAI’s new GPT-6.1 series.
The Bottom Line
Anthropic’s introduction of Claude Haiku 5.5 and associated price reductions for Sonnet 5.5 demonstrates the competitive pressure in the SI market to offer lower-cost, high-performance models. By introducing API credits and updated SDKs with beta support for computer and browser use, the company is expanding the accessibility and utility of its SI tools for developers and enterprises.