Claude Sonnet 5: Autonomous AI for Browser and Terminal

Claude Sonnet 5: A Major Leap in AI Autonomy

The latest innovation from Anthropic, Claude Sonnet 5, redefines our interaction with artificial intelligences by granting them unprecedented autonomy. Unlike previous versions, this AI can manipulate a browser and a terminal without continuous human intervention. This update marks a significant milestone in the evolution of AIs, where lengthy sequences of manual commands become obsolete. Claude Sonnet 5 is designed to perform tasks autonomously, freeing users from typical constraints.

The arrival of Claude Sonnet 5 on the market surprised many, both for its capabilities and its ease of use. Indeed, this model makes the use of browsers and programming accessible to a broader audience, previously limited to high-end models. This radical autonomy, promised by Anthropic, opens new perspectives for users who can now delegate complex tasks to the AI.

Autonomy and Innovation: The Winning Duo of Claude Sonnet 5

Claude Sonnet 5 stands out as the most autonomous model in the Sonnet series. It can plan actions, open browsers, and use terminals independently. This ability to manipulate various tools and pursue missions without human assistance represents a true technological breakthrough. Early users have described this innovation as a major advancement, particularly due to the AI's capability to verify and correct its own lines of code.

This autonomy is not limited to executing simple tasks. Claude Sonnet 5 can organize a series of actions to achieve a complex goal, which is particularly interesting for developers and businesses. Until now, such a level of autonomy was reserved for the highest-performing models. With Sonnet 5, Anthropic narrows the gap with models like Opus 4.8, offering comparable performance in evaluations dedicated to AI agents.

Tests have shown that Sonnet 5 outperforms its predecessor, Sonnet 4.6, in terms of reasoning, programming, and tool usage, making it an attractive choice for users looking to optimize their productivity.

An Attractive Offer but Under Surveillance

Anthropic has set a competitive launch price for Claude Sonnet 5, at $2 per million input tokens until August 31, 2026. After this date, the rate will increase to $3, and output tokens will rise from $10 to $15 per million. This attractive pricing makes Sonnet 5 accessible to a wide range of users while offering advanced features.

However, the issue of security remains crucial. Anthropic claims to have strengthened security measures to limit the risks of hallucinations and undesirable behaviors compared to Sonnet 4.6. The risks of hallucinations have significantly decreased compared to the previous version. This latest model also offers better resistance to manipulation attempts and fewer undesirable behaviors than Sonnet 4.6. Despite these improvements, Claude Sonnet 5 is not specifically trained for advanced cybersecurity tasks. During tests on developing exploits for Firefox, the AI failed to produce a functional exploit, highlighting its limitations in the most sensitive applications.

These results reassure security researchers but disappoint those hoping to use the AI for hacking tasks. Claude Sonnet 5 is now the default model for users of the Free and Pro offerings and is also available for Max, Team, and Enterprise subscriptions, as well as for Claude Code and the dedicated developer platform.