Anthropic's Claude Haiku 5.5 beats GPT-6 Luna, costs 75% less

Anthropic released Claude Haiku 5.5 a year after the previous Haiku update, positioning it as the low-cost, high-speed member of a line-up that also includes Fable/Mythos, Opus and Sonnet. The release represents the first time a Haiku model has been tuned across inference levels, and Anthropic is suggesting it be used together with Opus 5.5 or Sonnet 5.5 as a subagent rather than on its own. Alongside the launch, Anthropic halved the cache-read price for Claude Sonnet 5.5, suggesting it wants the cheaper model to draw more high-volume work.

The immediate test of the release is whether the price advantage translates into usage. Anthropic's own figures put Haiku 5.5 ahead of GPT-6 Luna on benchmark scores and around 75% cheaper to run than Haiku 4.5, and the Artificial Analysis Intelligence Index v4.3.2 score of 43 against 17 for the prior version points to a large step up. Whether developers treat those numbers as decisive for cost-sensitive work such as summaries, compactions and database queries is likely to shape how quickly the model is adopted. The choice may hinge less on raw benchmark wins than on how the model performs in real, high-volume tasks.