The whole network scolded Claude for becoming stupid, and the fate of Anthropic was revealed: It’s not the model that deceived you.

Does getting a bigger model equal smarter? [Introduction] Does changing to a larger model mean you are smarter? This may be the deepest misunderstanding among Claude Code users. Many people have switched all the way to the most expensive Fable for this reason. Recently, Anthropic personally clarified this misunderstanding. Have you ever had this kind of moment: Claude Code failed to write the code, and your first reaction was to quickly change to a stronger model. But this trick often doesn’t work and may even be a waste of money. Recently, Anthropic officially published a long article specifically talking about this matter. Anthropic official long article The reason is that too many people confuse the two options in the Claude Code: one is model selection (Model), and the other is effort (Effort). Anthropic official long article In the past, everyone’s understanding of these two options was very simple: changing to a larger model will make the AI ​​smarter; raising the Effort is nothing more than making the AI ​​think for a while. And just in March this year, this misunderstanding caused a lot of chaos. At that time, many developers discovered that the Claude Code in their hands suddenly "became stupid". The files that should be read are not read, the tests that should be run are not run, and the tasks are half-finished and they give up the task and ask you for more information. As a result, there was a lot of scolding on GitHub. The harshest blow came from Stella Laurenzo, AMD’s head of AI. She dug out the logs of 6852 sessions on GitHub. It was actually measured that Claude's amount of thinking dropped by 67% compared to two months ago. She concluded: Claude can no longer be trusted to do complex engineering work. Claude can no longer be trusted. At first, everyone thought that their prompt words were not written correctly, or that the configuration was wrong. After struggling for a long time, I suddenly realized that the problem was not with me at all, but that Anthropic quietly changed a setting. On March 4, in order to reduce delays, they lowered the default gear of the Effort option in Claude Code from high to medium.