AI company Anthropic is dropping two new Claude updates and a Computer Use feature. Claude 3.5 Sonnet and its smaller sibling Claude 3.5 Haiku claim serious performance strides from the company’s update. Claude 3.5 Sonnet gets Computer Use and better performance. 3.5 Haiku matches the performance of Sonnet predecessor Claude 3 Opus, at a similar speed and cost to the previous Haiku model.
If all of this is Greek to you, you obviously don’t code but that means you’re probably not going to let Anthropic’s AI agents take over your computer. That’s what Computer Use means, after all.
Claude 3.5 – now with AI autopilot
Claude 3.5 Sonnet was already a leader in AI coding but this update makes it even better with a small performance bump that claims to surpass competitors like GPT-4o and Gemini 1.5 across various math and reasoning benchmarks. The new version also introduces the Computer Use API, a current beta feature allowing the AI model to mimic direct human interactions with the desktop. The system can now execute commands like mouse clicks, keystrokes, and cursor movements by analyzing screenshots and determining where to move.
Anthropic describes it as an advanced automation system, basically robotic process automation (RPA) tools but more versatile. Several companies are already exploring the potential of this feature. AI-powered software developer Replit, for example, integrates the Computer Use API into its Replit Agent product to evaluate applications in real time. Other early adopters include Asana, Canva, and DoorDash.
Haiku is a smaller but still powerful model. Haiku outshines the previous generation’s Claude 3.0 Opus, scoring better than GPT-4o and the original Sonnet in coding tasks. Haiku is better at following instructions and at tool usage. It’s ideal for automation in user-facing products and generating personalized experiences from huge volumes of data—like purchase history, pricing, or inventory records.
To address potential risks like spam or misinformation, Anthropic has implemented classifiers to monitor how the API is used. The company assures users that human oversight remains essential, as actions are only taken based on specific prompts provided by users. This combination of enhanced performance and new automation tools positions Anthropic’s latest models as powerful and practical solutions for businesses.