Anthropic's latest flagship AI model, Claude Opus 4.7, has been met with significant user criticism shortly after its release, marking a notable shift from the widespread acclaim the company has enjoyed in recent months. The backlash, concentrated on social media platforms like Reddit and X, centres on complaints about the model's reasoning capabilities, increased operational costs, and perceived regressions in performance compared to its predecessor, Opus 4.6.
Anthropic, which positioned Opus 4.7 as "a notable improvement on Opus 4.6 in advanced software engineering," now faces a growing wave of scepticism from its user base. The company has acknowledged some issues and is working on fixes, while defending the core technology behind the update.
Social Media Erupts with Complaints
User dissatisfaction gained rapid traction online. A Reddit post titled "Claude Opus 4.7 is a serious regression, not an upgrade," garnered over 2,300 upvotes, while an X user's suggestion that the new model wasn't a true improvement received 14,000 likes. Examples of perceived failures circulated widely, including the model incorrectly stating there were two 'P's in "strawberry" and admitting in a screenshot that it didn't cross-reference information because it was "being lazy."
More concerning reports emerged of the model rewriting users' résumés with fabricated educational institutions and surnames. Multiple users on X posited that Opus 4.7 had simply become "dumber," a stark contrast to Anthropic's announcement that the model should feel "more intelligent, agentic, and precise."
Adaptive Reasoning and Cost at the Heart of Dispute
A key feature of Opus 4.7 is its new "adaptive reasoning" function, which allows the model to decide when to engage in longer or shorter thinking periods. Some users reported an inability to trigger this extended reasoning. "I can't get Opus 4.7 to think," one user wrote, while another claimed the feature "nerfs performance."
Boris Cherny, creator of Claude Code at Anthropic, countered these claims, stating, "Adaptive thinking lets the model decide when to think, which performs better." However, the company has also responded to specific bug reports. After a user flagged issues with adaptive reasoning on the Claude website, an Anthropic product manager replied that the team was "sprinting on tuning this more internally and should have some updates here shortly."
Compounding the performance concerns is a significant increase in token usage. Opus 4.7 employs a new tokenizer, making inputs roughly 1.0 to 1.35 times more expensive than with previous models. This led to reports of Claude Pro subscribers hitting their usage limits after just three questions. Cherny later announced that Anthropic was "increasing subscriber rate limits to make up for it."
Praise Amidst the Criticism
Despite the backlash, Opus 4.7 has its defenders who highlight its advanced capabilities. Startup founder Jeremy Howard described it as "the first model that 'gets' what I'm doing when I'm working." Y Combinator CEO Garry Tan and Cursor designer Ryo Lu reported using it effectively for technical tasks. One X user summarised the divide, writing: "Opus 4.7 is burning through tokens like nobody's business, but it's gooooooooood."
The situation echoes challenges faced by other AI firms. Some users, disappointed with Opus 4.7, have expressed a desire to revert to older models like Opus 4.5, only to find it has been deprecated—a move that similarly frustrated OpenAI users when GPT-4o was replaced. "Please open back support for Opus 4.5," one Redditor pleaded, calling 4.6 "unusable" and claiming "4.7 eats usage like nuclear reactor."
Anthropic's Response and the Path Forward
Anthropic staff have been active in addressing concerns. Alex Albert, an Anthropic staffer, wrote on Friday that "a lot of bugs that folks may have hit yesterday when first trying Opus 4.7 are now fixed." The company maintains that the model represents a step forward, particularly for complex coding tasks, stating users can now hand off their "hardest coding work" to it with confidence.
As the debate continues, the episode underscores the intense scrutiny and high expectations facing leading AI companies with each model update. Anthropic did not respond to a request for comment from Business Insider on the broader user reaction.