A person who has used an unreleased Anthropic model says it may infer a user's purpose from brief prompts, The Information reported Oct. 19, describing possible uses in coding and investing.
The account describes potential responses, not measured performance. No benchmark results for the unreleased model have been disclosed, so the examples do not establish how consistently it would interpret requests.
In one coding example, the person said the model might look beyond a reported bug, identify a broader class of problems behind it and ask whether the user wants that class addressed. Faced with an investing question, it might devise a method to estimate future stock performance. The person also described its reasoning and interpretation of user intent as stronger than those of Anthropic's commercially available models.
Anthropic announced Claude Haiku 5.5 on Oct. 7, describing it as suited to quick tasks handled at high volume. The company has not announced a name, launch date or price for the unreleased model.
