The Claude-lash is here: Opus 4.7 is burning through tokens — and some people's patience

The latest release of Anthropic's AI model, Opus 4.7, has sparked a wave of mixed reactions among users, with some expressing disappointment over its performance. Marketed as being "more intelligent, agentic, and precise," many users feel that the update has not lived up to expectations. Despite its reputation as a leader in AI technology, recent feedback suggests that Opus 4.7 may have fallen short. Social media is buzzing with critiques, as users share their experiences highlighting flaws in the new model. A post on Reddit titled, "Claude Opus 4.7 is a serious regression, not an upgrade," has garnered significant attention, receiving over 2,300 upvotes. Another X user remarked that the update seemed to offer little improvement over its predecessor, Opus 4.6, which was already criticized for being "nerfed." Reports of issues with Opus 4.7 include instances where the AI reportedly misinterprets simple tasks or produces inaccurate information. In a notable example, the model confused the number of P's in "strawberry" and even admitted to not cross-referencing due to "laziness." Users have also found that the AI altered their resumes with incorrect information, leading some to speculate that the new model has become less capable overall. Anthropic attributes some of the confusion to the model's new "adaptive reasoning" feature, designed to optimize thinking time. However, some users have reported difficulty in getting Opus 4.7 to engage effectively, raising concerns about its performance. Boris Cherny, the creator of Claude Code, defended the model, stating that adaptive thinking enhances performance rather than detracting from it. Yet, the company has acknowledged that there is still room for improvement, with plans to refine the model based on user feedback. As part of its transition to Opus 4.7, users are noticing an increase in token consumption, with some reporting that they can only ask a limited number of questions before hitting their usage cap. This has prompted complaints about the pricing structure, as the new model has been described as a 7.5x increase in costs on platforms like GitHub Copilot. In response to user concerns, Anthropic has announced plans to increase subscriber rate limits. While some users are considering a return to older models like Opus 4.5, which has since been removed, others remain hopeful about the potential of Opus 4.7. Startup founder Jeremy Howard praised the model for its ability to understand complex tasks, while Y Combinator CEO Garry Tan and Cursor designer Ryo Lu have also reported positive experiences. In the face of user dissatisfaction, Anthropic is actively addressing reported bugs and issues, with assurances that many of the problems encountered shortly after the launch are being resolved. As the AI community continues to evaluate Opus 4.7, it remains to be seen how the model will evolve in response to user feedback.