Anthropic has launched Claude Haiku 4.5, a compact artificial intelligence model designed to deliver high-performance capabilities at a fraction of the cost and with significantly greater speed compared to its larger counterparts.
The company asserts that Haiku 4.5 can perform comparably to models such as Sonnet 4, GPT-5, and Gemini 2.5, while costing one-third as much and operating at more than twice the speed. These claims aim to position the lightweight model as a powerful yet economical option for advanced AI deployments.
Internal tests cited by technology publication TechCrunch show Haiku 4.5 achieved 73% on the SWE-Bench and 41% on Terminal-Bench, a benchmark focused on command-line tasks. These results reportedly match those of Sonnet 4 in these specific metrics. Anthropic also reported similar performance across benchmarks for tool use, computer use, and visual reasoning.
Haiku 4.5 is immediately available across all free Anthropic plans. This move targets scenarios where server load and latency are critical, particularly for free AI product versions and large-scale deployments.
The model’s lightweight design facilitates the deployment of multiple agents in parallel or their integration with more complex models. Anthropic Chief Product Officer Mike Krieger stated that Haiku “opens up completely new categories of what is possible with AI in production environments.” He added that a model like Sonnet can handle complex planning while Haiku-powered sub-agents execute tasks rapidly.
Andrew Filev, CEO of Zencoder, described Haiku 4.5 as “unlocking a completely new set of use cases,” signaling strong commercial interest in the model’s capabilities.
Immediate applications are expected in software development tools, where low latency is crucial for developer workflows. Other potential uses include hybrid conversational assistants, automation pipelines, and task orchestration in cloud environments.
Industry analysts will closely monitor whether Haiku 4.5’s performance holds in independent evaluations and real-world production settings. TechCrunch noted that final comparisons will depend on external testing beyond Anthropic’s internal benchmarks.
The launch of Haiku 4.5 follows Anthropic’s recent releases of Sonnet 4.5 just weeks prior and Opus 4.1 approximately two months earlier. This consistent rollout reflects the company’s strategy of offering specialized models for diverse workloads and use cases. The previous version of Haiku was released in October 2024.