Some users have reported frustrations with Claude Opus 5, but a few simple setting changes can improve the AI experience.
Anthropic’s latest Claude Opus 5 has quickly become one of the most capable AI models on the market, but not everyone is enjoying the experience.
While the company describes it as a model that delivers near-flagship performance at roughly half the price of Claude Fable 5, some users say it overthinks simple tasks, produces unnecessarily long responses, and spends too much time verifying its own work.
“It’s a hard model to love,” wrote Every CEO Dan Shipper on X, while other users described it as “obtuse” and prone to making “smart-dumb mistakes” despite its impressive reasoning abilities.
Even so, benchmark results tell a different story. According to Artificial Analysis, Claude Opus 5 currently ranks first on its intelligence benchmark, outperforming nearly 190 AI models. The trade-off is speed. The model ranks much lower for response time and is significantly more verbose than many competitors.
Why does Opus 5 feel different?
Unlike previous Claude models, Opus 5 is designed to think more deeply before responding. It launches more internal sub-agents, checks its reasoning more frequently, and often verifies its answers without being asked.
That extra reasoning can improve performance on complex tasks, but it can also make straightforward requests feel slower and more expensive, particularly for API users who pay based on token usage.
Anthropic has acknowledged this behavior in its prompting guide and says a few small adjustments can noticeably improve the experience.
1. Lower the effort setting
One of the most effective changes is reducing the model’s Effort setting.
Effort controls how much reasoning Claude performs before generating a response. While higher settings can improve difficult problem-solving, many users have found that Low or Medium effort produces faster and cleaner answers for everyday tasks.
Anthropic’s own testing found that higher effort levels can sometimes cause the model to overthink problems or repeatedly reconsider answers. In fact, the company quietly updated its documentation shortly after launch, changing its recommendation from Extra High effort to simply High.
2. Ask Claude to be concise
Another common criticism is that Opus 5 explains too much.
Anthropic recommends addressing this directly by instructing the model to keep responses short unless additional detail is requested. A simple instruction in your system prompt or initial request can significantly reduce unnecessary explanations while also lowering token usage and cost.
3. Stop telling it to verify everything
Many experienced Claude users have long included prompts asking the model to double-check its work.
With Opus 5, that instruction is often unnecessary.
The model already performs considerably more internal verification than previous versions. Adding explicit verification prompts can cause repeated self-checking, longer responses, and unnecessary reasoning loops.
Anthropic suggests removing those instructions unless the task genuinely requires an additional layer of validation.
Is Opus 5 still worth using?
Despite the criticism, many AI experts continue to rate Opus 5 highly for complex work.
AI product expert Claire Vo called it “brilliant but annoying,” saying it consistently produced the highest-quality results in blind testing even though it was less enjoyable to work with.
Opus 5 also has advantages beyond pricing. Anthropic says it is significantly less restrictive than Claude Fable 5 on certain biology-related requests, blocking around 85% fewer prompts while maintaining its safety standards.
The bottom line
Claude Opus 5 is exceptionally capable, but its default behavior won’t suit everyone. Its tendency to overthink, generate lengthy responses, and verify its own reasoning can make simple tasks feel unnecessarily complicated.
For many users, lowering the effort level, asking for concise responses, and removing redundant verification instructions are enough to make the model noticeably faster, cheaper, and easier to use without sacrificing much of its reasoning ability.
Illustration: Inc; Photo: Adobe Stock, Anthropic
Source: Inc
