What's this about?
Anthropic released its fastest small model for summaries, classification, support, browser use and subagent work. Requests up to 100K tokens cost one tenth of Haiku 4.5; longer requests cross into a price tier five times higher.
What exactly is a small model?
It is the cheaper, faster member of a model family, chosen for frequent or latency-sensitive jobs rather than the hardest reasoning. An agent system can use a large model as the manager and send routine subtasks to Haiku.
What's the community saying?
launch post 37,381 likes and 1,260 replies; independent benchmark post 1,488 likes.
- Developers welcomed the $0.10/$0.50 price for ordinary prompts and the speed of subagents.
- Artificial Analysis found leading small-model intelligence but unusually high token consumption at maximum effort.
- Informal visual demos disagreed sharply on whether Haiku beats GPT-6 Luna, underscoring the weakness of cherry-picked examples.