— What was recently at the frontier is now cheaper and faster. Five months ago, Claude Sonnet 4 was a state-of-the-art model, writes Anthropic. — Today, Claude Haiku 4.5 gives you similar levels of coding performance, but at one-third the cost and more than twice the speed.
It really comes to life in Claude Code and in the API, where Sonnet 4.5 can make a multi-step plan for a complex problem and have multiple Haikus «compete subtasks in parallel.»
The model doesn’t top any benchmarks, coming in at 73.3% in SWE-bench Verified, but is slightly better overall than Sonnet 4, and even tops GPT-5 in some areas, particularly on subsets of agentic tool use.
Dirt cheap performance
What makes the model attractive is for people who want «near-frontier» levels of performance for a fraction of the price — the model costs only $1 per million input tokens, and $5 for outputs.
That makes it one of the cheapest almost-there models, but it is still topped by Grok 4 Fast at $0.20 and $0.50 for input/output.
Good for real-time tasks
Anthropic also pitches the model for use in real-time, low-latency tasks — like chatbots or customer service bots — for its speed and efficiency, and says it will make the coding experience «markedly more responsive.»
Haiku 4.5 is a drop-in replacement for Haiku 3.5, and is available now Anthropic’s apps and Claude Code. You can also try it out in the API, Amazon Bedrock and Google’s Vertex AI.
It will also be immideatly available under all of Anthropic’s free plans, where it will provide high performance for lower server loads — and, ultimately, cost, writes TechCrunch.
Read more: Anthropic’s launch page, writeups on TechCrunch, and Ars Technica.