Anthropic just dropped Claude Haiku 4.5, the latest version of its speed demon in the Claude family, and it’s already turning heads. The company claims Haiku 4.5 delivers performance comparable to Sonnet 4 but at one-third the cost and twice the speed. In other words, it’s fast, cheap, and surprisingly clever; a combo every engineer dreams of.

Benchmarks back the hype: Haiku scored 73% on SWE-Bench Verified and 41% on Terminal-Bench, roughly matching Sonnet 4, GPT-5, and Gemini 2.5. It also held its own across tool use, visual reasoning, and computer-use tests. For a model built to be lightweight, those are heavyweight numbers.

The kicker? Haiku 4.5 is immediately available for free under Anthropic’s plans. That means developers and startups can now build fast, low-cost agents without overloading their servers or budgets. Anthropic’s Chief Product Officer Mike Krieger says the release enables entirely new kinds of production deployments: “Sonnet handles the complex planning, while Haiku-powered sub-agents execute at speed.” In simpler terms, multiple intelligent agents can now run in parallel, maintaining performance without latency spikes.

Haiku 4.5 is expected to have an immediate impact in software development environments, especially within tools like Claude Code, where response times are crucial. Zencoder CEO Andrew Filev described the new version as “unlocking an entirely new set of use cases.” source

This launch follows a string of high-profile releases for Anthropic, just two weeks after Sonnet 4.5 and two months after Opus 4.1, both of which were hailed as state of the art. The previous version of Haiku was released in October 2024, showing that Anthropic is maintaining an aggressive release cadence.

Claude Haiku 4.5 proves that smaller models can now deliver big results, offering the perfect balance of intelligence, speed and cost for modern AI workloads.