Claude Haiku 5.5 arrives with massive price cuts proving the AI pricing arms race is far from over

Anthropic has released Claude Haiku 5.5, described as its fastest and most affordable small model, aimed at high-volume, cost-sensitive work such as summarization, database queries, classification, and live customer support. The model is available now on Amazon Web Services, Google Cloud, and Microsoft Azure.

Pricing is the headline change. Haiku 5.5 costs an average of about 75 percent less than Haiku 4.5, and for requests with prompts up to 100,000 tokens — which Anthropic says covered roughly 90 percent of previous Haiku requests — prices fall by up to 90 percent. Prompts beyond 100,000 tokens cost five times as much. Anthropic notes that Haiku 5.5 uses an updated tokenizer that consumes slightly more tokens per task than its predecessor, so real-world savings are likely smaller than per-token prices suggest; the same tokenizer effect raised usage about 30 percent on the Opus 4.x models.

Benchmark gains are large. Haiku 5.5 scores 1,620 on GDPval-AA v2.1, compared with 735 for Haiku 4.5. On Humanity's Last Exam it reaches 45.9 percent without tools and 57.4 percent with tools, up from 10.2 and 18.7 percent. Computer use shows the biggest jump: 72.4 percent on OSWorld-2.1 versus 15.7 percent. On the agentic coding benchmark Terminal-Bench 4.0 it scores 39.2 percent, while Haiku 4.5 scored zero. Anthropic lists OpenAI's budget model GPT-6 Luna as a comparison and says Haiku 5.5 leads in every tested category, though Sonnet 5.5 reference scores show it still trails Anthropic's larger model.

Haiku 5.5 is the first Haiku-class model with adjustable reasoning levels, letting users trade cost against quality. Anthropic recommends it for narrowly scoped tasks such as compaction, summarization, or sub-agent work, and says Sonnet 5.5 and Opus 5.5 remain better for complex agentic coding. Cybersecurity safeguards are tighter than the predecessor's but permit a broader range of defensive tasks than Sonnet 5.5, partly because the model is less capable overall; penetration testing stays blocked, and organizations can apply to Anthropic's life sciences and cybersecurity verification programs.

Anthropic is also halving Sonnet 5.5 cache read costs from $0.20 to $0.10 per million tokens, which it says should cut costs for most agentic tasks by about 20 percent. Monthly API credits are rolling out: $100 for Max-5x subscribers, $200 for Max-20x, and up to $500 for Team subscribers. The Python and TypeScript SDKs are being updated with beta support for computer use and browser use.

Why it matters: Developers running high-volume classification, summarization, or support workloads get a cheaper model with sharply higher computer-use and agentic coding scores, while Sonnet 5.5 users pay less for cached tokens. The updated tokenizer and the fivefold price for long prompts mean the effective savings depend on task length and token consumption.