Claude Haiku 4.5 is said to be Anthropic’s safest and fastest model to date.
Anthropic has introduced Claude Haiku 4.5, a new mini model that the company claims delivers powerful performance at lower costs. The model costs $1 per million input tokens and $5 per million output tokens. This makes it three times cheaper than the flagship Claude Sonnet 4.5.
Hybrid Reasoning Mode
Haiku 4.5 is a hybrid reasoning model. This means it decides for itself how much computing power is needed for a task. In standard mode, it uses little hardware and responds quickly. For more complex questions, the user can activate the extended thinking mode, where the model reasons more deeply and takes more time for its answer.
The model can process prompts up to 200,000 tokens and provide responses up to 64,000 tokens. This allows you to generate extensive business reports or technical documents, for example. According to Anthropic, Haiku 4.5 scores less than 10 percent below Sonnet 4.5 on most benchmarks, but outperforms older models in math and programming tests.
Focused on Efficiency and Safety
Anthropic calls Haiku 4.5 its safest and fastest model to date, claiming it to be more than twice as fast as Sonnet 4. According to Anthropic, the model can be deployed in applications where low latency is needed, such as customer services or AI agents.
Claude Haiku 4.5 is available via APIs, the Claude chatbot, and Claude Code. The latter is Anthropic’s programming assistant which, according to Reuters, is already generating nearly a billion dollars in annual revenue.