Anthropic, the company behind Claude, has addressed user complaints about its performance decline. The issue at hand is not that the model itself has become less capable, but rather that users have been misinterpreting how to use it effectively. This misunderstanding has led many developers to upgrade to more expensive models in an attempt to improve their AI’s intelligence, only to find that this trick does not always work as expected.
In fact, switching to a larger model is often not the solution to improving performance. According to Anthropic, users have been confusing two options in Claude Code: Model selection and Effort level. The default setting for Effort level was recently changed from high to medium, which has led many users to believe that their AI’s capabilities are declining.
The root cause of this issue is not a problem with the model itself but rather a misunderstanding about how to use it effectively. Many developers have been upgrading to more expensive models in an attempt to improve performance without realizing that the Effort level setting plays a crucial role in determining the AI’s behavior and output.
In March, many users reported issues with their Claude Code, including errors such as failing to read required files, skipping scheduled tests, and abandoning tasks halfway. The head of AI at AMD, Stella Laurenzo, conducted an investigation into these issues and found that Claude’s thinking volume had dropped by 67% compared to before February.
The issue was not with the model itself but rather a result of the default Effort level being changed from high to medium in March. This change was noted in the official update log, but many users did not notice it until later when they realized that their AI’s performance had declined.
After bearing the backlash for a month, Anthropic restored the default Effort level on April 7 and reset the usage quota for all subscription users. It was then that most users realized this switch had been right beside them all along, secretly determining whether the AI would devote its full capacity to work for them.
Anthropic’s official breakdown of how Claude works can be summed up in one simple sentence: Model determines capability, Effort determines dedication. The model defines the underlying capability and is backed by a set of ‘frozen weights’ that are permanently fixed during training. These weights cannot be altered or modified during inference, meaning that users can guide the AI but not train it.
Switching models essentially means replacing the entire set of weights to take over your task, solving the problem of whether the AI is capable of doing it. However, this does not necessarily mean that a larger model will always outperform a smaller one. In fact, Anthropic has found that a small model running at high Effort can absolutely outperform a large model running at low Effort on many tasks.
The division of responsibilities between Model and Effort is crucial to understanding how Claude works. The official judgment framework becomes extremely useful once users understand this concept. When Claude makes a mistake, the first step is always to check the context: is the prompt clear? Are all required tools provided? Is the CLAUDE.md file configured correctly?
Most so-called ‘AI getting dumber’ issues are rooted in incorrect configuration or insufficient dedication rather than an issue with the model itself. If the context is indeed correct but the AI still makes mistakes, users should ask themselves: is it incapable of doing the task, or is it just not trying hard enough? Not trying hard enough is easy to identify: it skips files that should be read, fails to run tests, and abandons a refactoring task halfway to ask for help.
If it’s genuinely ‘incapable’, then no amount of increased Effort will help. The issue lies with the model itself, and users need to switch to a more powerful model. Anthropic has given an analogy to explain this concept: Sonnet is like a versatile all-rounder who has an entire afternoon available to work; Opus is an expert who only gives you 5 minutes.
Each model has its strengths and weaknesses, and users should not treat it as a rigid token budget limit. The Effort setting in Claude Code determines how thoroughly and confidently the task needs to be completed, governing text responses, tool calls, and extended thinking all at once. For example, for the exact same prompt, Claude at high Effort can generate roughly 7 times more tokens than at low Effort.
The official team has also released an illustration showing that a small model running at high Effort can genuinely outperform a large model running at low Effort on many tasks. This is because a small model paired with sufficient context and high dedication can handle far more work than expected.
In the past, users would simply pick the most powerful model without considering other factors such as Effort level. However, this approach has become outdated, and users now need to act like project managers, assigning different roles and dedication levels to different models. This requires a new skillset: orchestrating AI agents rather than just relying on more powerful models.
The newly added ‘ultracode’ option in Claude Code’s Effort menu brings this orchestration mechanism directly into the product. When selected, Claude gets the xhigh Effort level, plus an authorization to decide on its own whether to spin up a team of AI agents, split the task, and run it in parallel.
The era of only looking at model rankings is passing, and orchestrating models is becoming a core skill. Whoever learns to assign tasks to AI first will get a head start, using a Claude that truly goes all out for them.