BitBytecGet early access
Illustration of a cloud above a swift paper-plane craft carrying coins and a microphone-like device emitting colourful sound waves.

AI ·

AWS adds Claude Haiku 5.5 and Nova 2.5 Sonic for faster AI agents

Amazon Web Services says Claude Haiku 5.5 costs about 75 percent less than its predecessor, while Nova 2.5 Sonic targets real-time voice agents.

News from

Picture: BitBytec · Original illustration, BitBytec

Two AI model launches stood out in Amazon Web Services' weekly news round-up of 13 October: Anthropic's Claude Haiku 5.5 and Amazon's own Nova 2.5 Sonic. AWS framed the week as "smaller, cheaper, and more constrained" [1]. Both models are aimed at businesses that want AI agents without paying top-tier prices.

Claude Haiku 5.5

AWS says Claude Haiku 5.5 is now available on Amazon Bedrock and Claude Platform on AWS, and that it costs around 75 percent less than Claude Haiku 4.5 for most tasks [1][2]. According to Anthropic, as relayed by AWS, it is the fastest and most efficient model in the Claude 5.5 family, built for subagents and high-volume, cost-sensitive work [2].

It is the first Haiku with effort controls, so a team can tune cost against intelligence for each task rather than choosing one setting for a whole workload [1][2]. AWS also says it is Anthropic's most capable Haiku across coding, tool use, computer use and agentic tasks [2].

The suggested pattern is a team of models. Claude Opus 5.5 plans and makes the judgment calls, while many Haiku subagents run in parallel for routing, classification and small code changes across many files [1][2]. AWS lists other uses such as summarising documents, answering quick questions over a knowledge base and repetitive browser and desktop tasks [2].

For companies already on AWS, the platform keeps data inside AWS infrastructure with regional data residency, works with existing access, audit and monitoring controls, and usage appears on the AWS bill [2]. AWS's dedicated announcement is dated 7 October [2].

Amazon Nova 2.5 Sonic

Nova 2.5 Sonic is a speech-to-speech model for building real-time voice agents. AWS says it offers better reasoning, instruction following and tool-calling accuracy, plus lower latency [1]. Its example is a support agent that looks up an order, checks return eligibility and starts the return within one conversation [1].

The model keeps expressive voices in seven languages, controllable turn-taking and a 256K context window, and pairs with Strands Bidi Agents, which AWS says are now generally available, to build a production voice agent in a few lines of code [1]. AWS gave no price or benchmark figures for Nova 2.5 Sonic in the round-up.

Also in the round-up

AWS also introduced Strands Box, an open-source developer-preview sandbox that combines operating-system containment with fine-grained policies, so an agent can be limited to actions such as posting to an incident channel a set number of times [1]. It fits the theme of putting boundaries around what agents may do.

What it means for business

The pitch from AWS is that agent work no longer needs the biggest model for every step. A cheaper, faster model handling routine volume while a stronger one plans could cut running costs, and the effort controls give finance and engineering teams a dial to manage spend. Voice agents for customer support are the other obvious use. The 75 percent figure applies "for most tasks" and comes from AWS and Anthropic rather than independent tests, so run a pilot on your own workloads before changing budgets.

Source: AWS — AWS Weekly Roundup: Prime Day 2026 by the numbers, Claude Haiku 5.5, Amazon Nova 2.5 Sonic, and more (October 13, 2026)

  • aws
  • amazon bedrock
  • claude haiku
  • amazon nova
  • voice agents
  • anthropic

Sources

  1. AWS Weekly Roundup: Prime Day 2026 by the numbers, Claude Haiku 5.5, Amazon Nova 2.5 Sonic, and more (October 13, 2026) — AWS News Blog (primary source)
  2. Introducing Claude Haiku 5.5 on AWS — AWS Machine Learning Blog

Written with AI from the sources listed, checked against them before publishing.

Your idea. Your AI. Your creation.

What would you make first?

Describe it to the chat you already use. Accept the plan on your Mac. Watch your own coding agent build it.