Loading market data...

Anthropic Updates Claude 4.6 and Opus 4.7 With Scaling Tips, Accuracy Fixes for Browser Tasks

Anthropic Updates Claude 4.6 and Opus 4.7 With Scaling Tips, Accuracy Fixes for Browser Tasks

Anthropic has released updates to its Claude 4.6 and Opus 4.7 models, adding scaling tips and accuracy fixes specifically for browser and computer use tasks. The changes target how the AI handles actions like clicking, scrolling, and filling forms, aiming to reduce errors that can trip up automation workflows.

What the updates change

The new versions include modifications to Claude's underlying inference logic rather than major architectural overhauls. Scaling tips are intended to help the model better judge when to break a broad instruction into smaller steps, while the accuracy fixes address common missteps in interpreting web-page coordinates or button labels. Anthropic says the improvements make the models more reliable for repetitive tasks such as data entry, form submission, and multi-step web navigation.

Why browser and computer use matter

Claude's ability to interact with graphical interfaces — not just text — has been a growing focus for Anthropic. Businesses and developers use the model for robotic process automation, customer support scripts, and software testing. Errors in those contexts can cascade quickly: a misclicked button or a wrongly parsed dropdown can break an entire sequence. The company has been iterating on these capabilities since introducing computer-use features in late 2024.

Who gets the updates

The 4.6 and Opus 4.7 versions are available to API users and through Anthropic's console. Opus 4.7 is the more powerful tier, aimed at complex, high-stakes tasks; 4.6 is positioned as a faster, lighter option. Both models now ship with the new scaling tips and accuracy fixes enabled by default, according to the company's release notes.

The changes come as rivals like OpenAI and Google push similar agent-style features. For now, Anthropic is focusing on incremental reliability gains rather than flashy new abilities. Developers who rely on Claude to drive browser-based workflows will get the updates automatically — no need to reconfigure prompts or retrain the model. The fixes are already live in production.