Anthropic launches Claude Haiku 4.5 for low‑latency, low‑cost AI

Anthropic has unveiled Claude Haiku 4.5, the latest version of its smallest large language model, positioning it as a low‑latency, cost‑efficient option that has matched or surpassed the company’s earlier mid‑tier Claude Sonnet 4 on several tasks while coming in at roughly a third of the price.

The model has been made available immediately across Anthropic’s apps and developer tools, as well as via Amazon Bedrock and Google Cloud’s Vertex AI.

What does Haiku do?

In launch materials and press briefings, Anthropic has positioned Haiku 4.5 as a “near‑frontier” model built for real‑time chat, customer support agents and pair programming, with more than double the speed of its predecessor and performance comparable to Claude Sonnet 4 on benchmarks that include coding and computer use.

Anthropic’s Chief Product Officer Mike Krieger has said the release has enabled economical deployment of multiple sub‑agents at scale.

Anthropic has also positioned Haiku 4.5 as substantially more affordable than its higher‑end models: about one‑third the cost of Sonnet 4 and roughly one‑fifteenth the cost of Opus, according to the company’s briefings to media.

Investing.com reported list API pricing at $1 per million input tokens and $5 per million output tokens under the “claude‑haiku‑4‑5” identifier.

Availability on AWS and Google Cloud

Amazon Web Services confirmed that Claude Haiku 4.5 has been added to Amazon Bedrock with global cross‑region inference options, aimed at latency‑sensitive, production‑scale workloads.

Google has listed the model as generally available on Vertex AI (version “claude‑haiku‑4.5@20251001”) with a 200,000‑token context window and regional endpoints in the US, Europe and Asia Pacific.

What has changed?

• Speed: Anthropic has said responses are more than twice as fast as earlier Haiku releases, making it suitable for live assistants and call‑centre workflows.
• Capability: Company claims indicate parity with Sonnet 4 on several benchmarks, including coding and “computer use” tasks, with improved tool and agent orchestration.
• Deployment: Immediate availability across Anthropic apps, Claude Code and partner clouds (AWS Bedrock, Vertex AI) has broadened routes to production.

Anthropic also added Haiku 4.5 has demonstrated low rates of concerning behaviours in internal tests and has been released under its AI Safety Level 2 (ASL‑2) classification—less restrictive than the ASL‑3 designation applied to Sonnet 4.5 and Opus 4.1.

The launch has arrived amid rapid enterprise uptake of Anthropic’s stack. Reuters reported that Anthropic has served more than 300,000 business customers and has been approaching a $7 billion annual revenue run‑rate, with the company emphasising cost‑effective deployments to mainstream industries.