Claude Haiku 5.5 debuts with steep price cuts, underscoring that the AI pricing arms race is far from over

Meta description: Anthropic launches Claude Haiku 5.5, cutting small-model costs by up to 90 percent, and lowers Sonnet 5.5 cache read prices amid benchmark gains.

Tags: Claude Haiku 5.5, Anthropic AI, Sonnet 5.5, AI benchmarks, agentic coding

Featured image: abstract technology style with flowing data streams and neural network motifs; no real people, company logos, or text.

Anthropic Claude Haiku 5.5 Launches as Sonnet 5.5 Prices Fall

Anthropic has released Claude Haiku 5.5, positioning it as the company's fastest and most affordable small model to date. The launch matters because it pairs a large benchmark improvement with a steep price cut for high-volume workloads, while Sonnet 5.5 also becomes cheaper. Claude Haiku 5.5 is aimed at tasks where speed and unit economics decide whether a model can be used widely, from summarization and database queries to classification and live customer support.

Before the release: the small-model cost squeeze

Haiku-class models are used in places where many calls happen at low individual value. A small price difference can therefore change what teams can afford to run continuously. Anthropic had previously sold Haiku 4.5 into that role, but its benchmark results showed clear limits. The company also faced a market in which lower-cost AI systems are becoming more competitive, as shown by OpenAI's budget model GPT-6 Luna and the newer GPT-6.1 series.

The launch response is therefore not only about capability. It is about whether a small model can now handle more work without forcing users to move to a larger, more expensive model.

What changed at launch: benchmarks, pricing, and tokenizer caveats

At release, Anthropic reported a major performance jump. On the knowledge benchmark GDPval-AA v2.1, Claude Haiku 5.5 scores 1,620, more than double the 735 managed by 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.

The biggest jump appears in computer use, where the model operates a computer on its own. This workload burns through large amounts of tokens, making a cheap, fast model such as Claude Haiku 5.5 a natural fit. It scores 72.4 percent on OSWorld-2.1, up from 15.7 percent. On the agentic coding benchmark Terminal-Bench 4.0, it reaches 39.2 percent, while Haiku 4.5 scored zero.

Anthropic also lists OpenAI's budget model GPT-6 Luna as a comparison, and Claude Haiku 5.5 leads across every tested category. Sonnet 5.5 reference scores show, however, that Claude Haiku 5.5 still falls well behind Anthropic's larger model.

Pricing is the other main change. On average, Claude Haiku 5.5 costs about 75 percent less than Haiku 4.5. For requests with prompts up to 100,000 tokens, which Anthropic says account for roughly 90 percent of all previous Haiku requests, prices drop by up to 90 percent. Prompts longer than 100,000 tokens cost five times as much.

Anthropic also points out that Claude Haiku 5.5 uses an updated tokenizer, a component that converts text into numeric tokens for processing. It consumes slightly more tokens per task than its predecessor. The same thing happened with the Opus 4.x models, where token usage jumped about 30 percent from the tokenizer change alone. That means real-world savings are likely smaller than the per-token prices suggest.

First adjustable reasoning in a Haiku model

Claude Haiku 5.5 is the first Haiku-class model with adjustable reasoning levels, letting users balance cost against quality. Anthropic says the model works best for narrowly scoped tasks like compaction, summarization, or sub-agent work. For complex agentic coding, Sonnet 5.5 and Opus 5.5 remain the better picks.

Cybersecurity safeguards are tighter than on the predecessor but allow a broader range of defensive tasks than Sonnet 5.5, partly because Claude Haiku 5.5 is less capable overall. Penetration testing stays blocked. Organizations with broader needs can apply for Anthropic's verification programs for life sciences and cybersecurity.

Latest progress: availability, Sonnet price cuts, and SDK beta features

Claude Haiku 5.5 is available now across all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Alongside the Haiku launch, Anthropic is cutting cache read costs for Sonnet 5.5 by 50 percent, from $0.20 to $0.10 per million tokens. The company says this should reduce costs for most agentic tasks by about 20 percent. The change arrives as OpenAI's new GPT-6.1 series intensifies AI price competition.

The company is also handing out API credits and updating its Python and TypeScript SDKs, adding beta support for computer use and browser use.

What to watch next is whether actual tokenized task totals, not sticker prices, determine real savings; how teams use adjustable reasoning levels; and whether Sonnet 5.5 cache cuts become a longer-term pricing trend rather than a one-off adjustment.