Anthropic's Claude Haiku 5.5 cuts prices 75%, scores 1,620

Anthropic positions Haiku 5.5 as its fastest and most affordable small model to date, aimed at high-volume, cost-sensitive work such as summarization, database queries, classification, and live customer support. The company says the average cost is about 75 percent below Haiku 4.5, and for prompts up to 100,000 tokens — which Anthropic says cover roughly 90 percent of all previous Haiku requests — prices drop by up to 90 percent.

The benchmark picture is where the release gets its edge. Haiku 5.5 scores 1,620 on GDPval-AA v2.1, more than double Haiku 4.5's 735, and it jumps from zero to 39.2% on Terminal-Bench 4.0. Computer use shows the largest gain, reaching 72.4% on OSWorld-2.1 versus 15.7% — a fit, the company notes, because computer use burns through large amounts of tokens. Haiku 5.5 is also the first Haiku-class model with adjustable reasoning levels, letting users trade cost against quality, though Anthropic still points complex agentic coding toward its larger models.

One caveat cuts against the headline numbers: Anthropic says the updated tokenizer consumes slightly more tokens per task, as happened with the Opus 4.x models where token usage jumped about 30 percent from the tokenizer change alone. Real-world savings therefore are likely to be smaller than per-token prices suggest. How much of the promised cost cut survives that effect — for teams running summarization or support agents at scale — is the practical test.