Anthropic is reportedly moving Claude toward a more independent way of working. According to The Information, the company plans to release new versions of Claude Opus and Sonnet in the coming weeks, with testers describing models that can take more initiative during difficult tasks.
The core idea is simple but important: instead of waiting for a user to redirect them whenever something goes wrong, the new Claude models are said to analyze the problem, adjust their approach, and keep working.
What Anthropic Is Reportedly Testing
The reported update focuses on greater autonomy and self-correction. Testers say the new Claude models can operate much more independently than earlier versions, especially when a task requires both reasoning and external tool use.
That combination matters because many useful AI tasks are not solved in one straight line. A model may need to gather information, test an idea, realize the first path is weak, and then choose a better route. The Information describes new Claude versions that can move between those modes more smoothly.
If a model gets stuck while using a tool, it can enter a "thinking" mode to examine what happened. From there, it can attempt to fix the issue instead of simply stopping or asking the user for a new instruction.
Why Self-Correction Changes the Workflow
Self-correction is not just about catching a typo or retrying a failed command. In the examples described, the model is expected to judge whether its current approach is useful for the problem in front of it.
One example from The Information involves a market analysis for a Manhattan café. The model first looks at national trends, then recognizes that those trends are not useful enough for the specific question. It shifts toward demographic data from the East Village to produce recommendations that better fit the café’s local context.
That kind of adjustment is the practical value of the reported Claude update. The user gives a goal, and the model has more room to decide which information is relevant, when an answer is drifting off target, and how to recover.
For users, the result could be fewer step-by-step corrections. Instead of repeatedly explaining why a path is not useful, they may be able to describe the outcome they want and let the model handle more of the intermediate judgment.
Coding Tasks Could See the Biggest Impact
The new Claude models are also reportedly more active when working with code. According to the source article, they can automatically test the code they generate. If something fails, they can pause, inspect the problem, and try to repair it themselves.
That is a different workflow from simply producing code and leaving all validation to the user. A model that tests its own output can catch some problems before the user sees the result. It can also use failures as feedback, which makes the coding process more iterative.
Early testers say this behavior can apply even when the prompt is broad. With an instruction such as "make the app faster," the model may independently try different optimization strategies rather than asking for a narrow technical checklist.
The important point is not that the model becomes perfect. The source does not claim that. The important point is that Anthropic is reportedly trying to make Claude more capable of continuing work when the task is open-ended, unclear, or technically messy.
Part of a Wider Shift in AI Models
Anthropic’s reported direction fits a broader push toward AI systems that need less constant steering. The updated Claude models are designed to combine reasoning and tool use, switching between them as the task demands.
The source article compares this direction with OpenAI’s o3 and o4-mini models. Earlier o1 models could "think through" extra steps by generating text. The newer generation can also use tools such as web search, generate code, or analyze images as part of its reasoning.
That added tool access is meant to make models more flexible and robust. At the same time, the source notes that initial tests show o3, for example, still makes mistakes on complex tasks more often than previous OpenAI models.
That caveat is important for understanding the stakes. More autonomy can make AI systems more useful, but it also raises the importance of reliability. A model that acts with less guidance needs to be good at recognizing bad paths, correcting errors, and knowing when its work is not yet good enough.
What to Watch Next
The reported Claude Opus and Sonnet updates point toward AI tools that behave less like passive text generators and more like systems that can manage a task through several stages. They may research, reason, use tools, test outputs, and revise their own work.
For businesses and developers, the appeal is clear: fewer interruptions, more initiative, and better handling of complex requests. For users, the main question will be whether that independence produces better results in practice.
Anthropic is reportedly preparing the next generation of Claude for release in the coming weeks. If the tester reports hold up, self-correction may become one of the most important measures of progress in everyday AI work.