AI News · Models ·
Anthropic launches Claude Haiku 5.5 with lower pricing
Anthropic launched Claude Haiku 5.5, the third model in its Claude 5.5 family introduced in the past month. Reuters reports that it costs 75% less than its predecessor, Haiku 4.5, with input pricing of $0.10 per million tokens. The model targets classification, summarization and extraction, including applications in live support, voice agents and in-app assistants.
Key points
- Haiku 5.5 is Anthropic’s third Claude 5.5 model launched in the past month.
- Reuters reported pricing 75% below Haiku 4.5, with input rates starting at $0.10 per million tokens.
- Higher input and output rates apply to longer prompts.
- Target uses include classification, summarization, extraction, live support, voice agents and in-app assistants.
- Anthropic said new safeguards cover a narrow set of high-risk cybersecurity requests.
What happened: Anthropic has launched Claude Haiku 5.5, an artificial intelligence model aimed at tasks including classification, summarization and extraction, Reuters reported. It is the third model in the Claude 5.5 family introduced in the past month, expanding the company’s lineup ahead of a planned initial public offering. For business teams, the central change is pricing: Reuters reported that Haiku 5.5 costs 75% less than its predecessor, Haiku 4.5.
The details: Pricing depends on prompt length, with separate rates for input and output. For prompts under 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. For longer prompts, those rates rise to $0.50 and $2.50, respectively. The headline input price therefore does not apply to every workload. Teams assessing the model need to consider both sides of the pricing schedule, rather than treating the lowest advertised rate as the full cost of use.
Who it affects: The model’s target applications include live support, voice agents and assistants embedded in apps, according to Reuters. Those are the use cases named alongside its focus on classification, summarization and extraction. For teams running these tasks at high volume, the lower price makes Haiku 5.5 a candidate for comparison with their current models. However, comparative task accuracy, response speed and results from business deployments were not reported, leaving important questions about whether it would be a suitable replacement.
The details: Anthropic said Haiku 5.5 is its first Haiku model with built-in safeguards for a narrow set of high-risk cybersecurity requests. The company added that most everyday tasks would not be affected. Reuters did not report which specific requests fall within that restricted set or how the safeguards work. The announcement therefore establishes a change in the model’s protections, but does not give teams enough detail to assess their effect on particular cybersecurity workflows.
What to watch: The practical next step is to benchmark Haiku 5.5 against existing models on the tasks a team actually runs. That comparison should test whether the lower pricing translates into savings without sacrificing task accuracy, while accounting for the higher rates on longer prompts. The launch offers a clear pricing change; whether it improves the economics of a particular workflow remains to be established through that evaluation.
Our take
Teams running high-volume workflows should benchmark Haiku 5.5 against their current models to see whether the lower price translates into savings without sacrificing task accuracy.