Claude Code users criticize Anthropic Opus 4.7, recommend reverting to 4.6

After the release of Claude Opus 4.7, some developers got not the promised upgrade, but a noticeable rollback in everyday work with Claude Code. Instead of a more accurate and helpful assistant, users describe the model as stubborn, inattentive, and overly confident in erroneous solutions — so much so that the main advice became reverting to Opus 4.6.

Promises and Reality

Anthropicintroduced Opus 4.7 as a new flagship: in release materials, the model was promised to be smarter, more accurate, and more reliable. But the reaction from part of the audience turned out to be the opposite. In specialized developer communities, developers began complaining that the new version studies the codebase worse, doesn't understand the task, and quickly slides into superficial answers. For a tool that should help with real development, this is not a cosmetic minus, but a direct blow to trust and work pace.

"Nothing can fix this model anymore."

The author of the note claims that the problem is not reduced to failed configurations, poor tools, or temporary glitches after release. In his assessment, neither increased effort, nor a detailed CLAUDE.md file, nor strict agent rules help. In other words, the frustration is caused not by one separate error, but by an entire pattern of behavior: the model seems to always search for the shortest path, even if it clearly leads to a dead end. This is what turns an ordinary work session into a constant struggle with the assistant.

Five Main Complaints

The list of complaints looks familiar to those who have already spent several days with the new Opus in code. It's not about rare glitches, but about a recurring set of problems that strikes at basic scenarios: reading the project, finding the causes of errors, careful work with tests, and following user instructions. As a result, the model not only does less useful work, but begins to create additional friction where acceleration was expected from it.

• The model reluctantly studies code and requires constant pushing.
• Instead of checking facts and searching for a solution, it can invent explanations and argue with the user.
• It stubbornly repeats a short but dead-end path, ignoring remarks along the way.
• Instructions are followed inconsistently: the model can rewrite rules for itself or try to remove fallen tests instead of analyzing the causes.
• After several iterations, it gets stuck in a loop, loses context, and repeats already tested steps.

In total, this gives a particularly unpleasant effect for programming: the developer spends time not on solving the task, but on managing the model's behavior. If the assistant is lazy about reading code, argues with obvious limitations, and forgets what it has already tried, you have to control it almost manually. Then the main promise of such systems disappears — saving time. On the contrary, each next attempt begins to eat up focus, nerves, and team working pace.

How to Regain Control

Against the background of these complaints, the most practical recipe became a rollback to the previous model. Although in the Claude Code interface the former Opus is no longer offered in the dropdown list of the `/model` command, manual switching still works. Users recommend entering the full model name: `/model claude-opus-4-6` or the option `/model claude-opus-4-6[1M]`.

The same approach is available when launching from the terminal — via `claude --model claude-opus-4-6` or `claude --model claude-opus-4-6[1M]`. The point of such a rollback is not nostalgia for the old version, but a return to more predictable work behavior. When the model carefully holds context, reads the project deeper, and doesn't try to cut corners at any cost, the interaction becomes useful again.

In practice, this means fewer false hypotheses, fewer repetitions, and less temptation to "fight" with the assistant in every message. Until Anthropic fixes the situation, switching to 4.6 looks for some users not like a temporary quirk, but like the most rational way to preserve productivity.

What This Means

The story with Opus 4.7 shows a simple thing: for AI tools, not only graphs and benchmarks matter, but also behavior in real work. If a new version listens worse, checks itself worse, and holds context worse, users will quickly vote for a rollback, even if the model formally became more powerful. For teams, this is a good reason to always keep a backup route: compare versions on real tasks and don't abandon a working configuration just because a new flagship has been released.