Anthropic launches Claude Haiku 5.5 with lower token prices

Anthropic has released Claude Haiku 5.5, a new small model that the company positions as its fastest and lowest-cost Haiku option so far. The launch matters for teams using AI in everyday workflows because it pairs lower token pricing with stronger benchmark results, especially for high-volume tasks such as summaries, classification, data queries and customer support.

For creators, marketers, freelancers and small businesses, the headline is not just that another model has arrived. It is that the cheapest tier of Claude is becoming more viable for routine automation where every API call has to justify its cost.

What changed in Claude Haiku 5.5

According to Anthropic, Haiku 5.5 is built for cost-sensitive jobs that need speed and scale rather than the full capability of the company’s larger models. The company names summarization, database queries, classification and live customer support as example use cases.

The pricing shift is the main commercial change. Anthropic says Haiku 5.5 costs about 75 percent less than Haiku 4.5 on average. For prompts up to 100,000 tokens, which the company says represented about 90 percent of earlier Haiku requests, token prices fall by up to 90 percent. Prompts above 100,000 tokens are priced five times higher than the lower tier.

The listed per-million-token prices for Haiku 5.5 are:

• Cache reads: $0.01 for prompts up to 100,000 tokens, or $0.05 above that
• Cache writes: $0.125, or $0.625 above 100,000 tokens
• Input tokens: $0.10, or $0.50 above 100,000 tokens
• Output tokens: $0.50, or $2.50 above 100,000 tokens

There is an important caveat for anyone calculating savings. Anthropic says Haiku 5.5 uses an updated tokenizer that consumes slightly more tokens per task than Haiku 4.5. The Decoder notes that a similar tokenizer change in Opus 4.x increased token use by about 30 percent, so real-world savings may be smaller than the headline token-price cuts suggest.

Benchmark gains point to broader everyday use

Anthropic’s benchmark data shows Haiku 5.5 moving well beyond Haiku 4.5 in several categories. On GDPval-AA v2.1, a knowledge work benchmark, Haiku 5.5 scores 1,620 versus 735 for Haiku 4.5. On Humanity’s Last Exam, it reaches 45.9 percent without tools and 57.4 percent with tools, compared with 10.2 percent and 18.7 percent for the previous Haiku model.

The largest jump cited is in computer use. Haiku 5.5 scores 72.4 percent on the offline subset of OSWorld 2.1, up from 15.7 percent for Haiku 4.5. Anthropic also reports 39.2 percent on Terminal-Bench 4.0, an agentic coding benchmark where Haiku 4.5 scored 0.0 percent.

Those results matter because computer use and agent-style workflows can consume large volumes of tokens. A cheaper model that can handle narrower computer-use tasks may let developers reserve larger models for the parts of a workflow that need more reasoning.

Anthropic also compares Haiku 5.5 with OpenAI’s GPT-6 Luna and says Haiku leads across the tested categories shown in its table. At the same time, Anthropic’s own Sonnet 5.5 remains ahead in the reference scores, including on OSWorld 2.1, Terminal-Bench 4.0 and Humanity’s Last Exam.

Where it fits for businesses and automation builders

Haiku 5.5 is the first Haiku-class model with adjustable reasoning levels, according to Anthropic. That gives API users a way to trade off cost and quality depending on the job.

For small teams, that makes the model most interesting in repeatable workflows: cleaning up text, compressing context, classifying inbound requests, drafting short summaries or powering sub-agents inside a larger process. Anthropic says Haiku 5.5 works best on narrowly scoped tasks such as compaction, summarization and sub-agent work.

The same guidance suggests clear limits. For complex agentic coding, Anthropic says Sonnet 5.5 and Opus 5.5 remain the better choices. In practice, that means Haiku 5.5 looks more like a lower-cost worker for high-volume background tasks than a universal replacement for larger models.

If you are comparing chatbot platforms for business use, this update also reinforces a wider pattern: model selection is becoming less about choosing one assistant and more about matching model size, price and task type. Our ChatGPT vs Claude vs Google Gemini buying guide covers that broader decision for teams choosing between major AI assistants.

Availability, credits and safeguards

The Decoder reports that Haiku 5.5 is available now across all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic is also reducing Sonnet 5.5 cache read costs by 50 percent, from $0.20 to $0.10 per million tokens. The company says that should lower costs for most agentic tasks by about 20 percent.

Alongside the pricing changes, Anthropic is adding monthly API credits for subscribers. Max-5x subscribers get $100, Max-20x subscribers get $200 and Team subscribers receive up to $500 per month, according to the company. Those credits can be used for experimenting with tools, apps and agents through the API.

The company is also updating its Python and TypeScript SDKs with beta support for computer use and browser use.

On safety, Anthropic says Haiku 5.5 has tighter cybersecurity safeguards than its predecessor while allowing a wider range of defensive tasks than Sonnet 5.5, partly because Haiku is less capable overall. Penetration testing remains blocked, while organizations with broader requirements can apply for Anthropic’s verification programs for life sciences and cybersecurity.