For a while it felt like Anthropic had forgotten about Haiku.
The last Haiku model came out almost a year ago, and since then all the attention went to the big models: Fable, Mythos, Opus and Sonnet. And honestly, those models are really good, especially Opus 5.5. So people slowly stopped talking about Haiku too.
Then Anthropic did it.
On October 7, 2026, they launched Claude Haiku 5.5.
The biggest headline is the price, but that’s not the only story. In this article we’ll look at:
How much cheaper it actually is (and the catch)
How much it improved over Haiku 4.5
How it compares to its biggest rival, OpenAI’s GPT-6 Luna, plus Gemini and DeepSeek
The 10 best use cases, so you know what it’s really capable of
slapping “10x cheaper” on the headline is marketing until you show the cost per successful task. show the pricing source, the same test prompts, and how many retries each model needed. cheap tokens get expensive fast when your app keeps paying to fix bad answers
Interesting analysis. What stands out most is the shift in the cost-to-performance ratio. If Haiku 5.5 truly delivers near-premium capabilities at a fraction of the cost, it could become the default choice for startups, developers, and high-volume…
