Anthropic Unveils Upgraded Claude AI Models and Experimental ‘Computer Use’ Feature
The enhanced Claude 3.5 Sonnet, new Claude 3.5 Haiku, and cutting-edge AI-driven computer use are set to transform multi-step automation and coding tasks.
Introducing Claude 3.5 Sonnet and Claude 3.5 Haiku
Anthropic has announced the launch of two new AI models: the upgraded Claude 3.5 Sonnet and the all-new Claude 3.5 Haiku. The refreshed Claude 3.5 Sonnet comes with significant improvements across all areas, particularly excelling in coding tasks, an area where it has consistently led the field. Claude 3.5 Haiku, on the other hand, rivals the performance of the previous Claude 3 Opus, providing similar speed and cost efficiency.
Revolutionary ‘Computer Use’ Capability
Alongside these model upgrades, Anthropic has introduced an experimental capability called “computer use,” now available in public beta. This feature allows developers to instruct Claude to interact with computers much like humans—by navigating screens, moving cursors, and inputting data through typing. While still in an experimental phase and prone to some errors, the introduction of this functionality marks a major step forward in AI development.
Early Applications of ‘Computer Use’
Several companies, including Asana, Canva, Cognition, DoorDash, Replit, and The Browser Company, are already experimenting with the “computer use” feature. For instance, Replit is utilizing Claude 3.5 Sonnet’s abilities in computer navigation to create a tool for evaluating apps as they are being developed within their Replit Agent product.
Claude 3.5 Sonnet: Leading in Software Engineering
The updated Claude 3.5 Sonnet shows significant gains in agentic coding and tool use tasks. It has notably improved its performance in coding benchmarks, such as increasing from 33.4% to 49.0% on the SWE-bench Verified task. It has also surpassed other models, like OpenAI o1-preview, in specialized agentic coding systems.
Customer feedback indicates that Claude 3.5 Sonnet has greatly enhanced AI-powered coding, offering stronger reasoning and no added latency, making it a preferred choice for multi-step software development tasks. Companies like GitLab and Cognition have reported notable improvements using the model for DevSecOps tasks and autonomous AI evaluations.
Claude 3.5 Haiku: Speed, Affordability, and Performance
Claude 3.5 Haiku represents the next generation of Anthropic’s fastest models, delivering better performance across the board. With enhanced coding skills and low latency, it outperforms many state-of-the-art models in intelligence benchmarks. Claude 3.5 Haiku is ideal for user-facing products, sub-agent tasks, and handling vast amounts of data for personalized experiences.
This model will be available later in the month via Anthropic’s API, Amazon Bedrock, and Google Cloud’s Vertex AI, starting as a text-only model, with image input support to follow.
Teaching AI to Navigate Computers
The “computer use” feature represents a new direction for AI. Instead of creating specific tools for Claude to complete individual tasks, the goal is to teach it general computer skills that can be applied to a range of tools and software. This new capability allows developers to automate repetitive tasks, conduct research, and interact with data in new ways.
With this feature, Claude can follow user instructions and translate them into commands like moving cursors, navigating web pages, or filling out forms. Despite some current limitations, early testing shows promising results, with Claude outperforming other AI systems in computer interface tasks.
Looking Ahead
Anthropic is continuing to refine the computer use feature and improve its AI models based on developer feedback. The Claude 3.5 models and this new capability mark a significant leap forward in AI development, offering new possibilities for multi-step automation and software engineering. As these technologies evolve, they are expected to unlock further innovation in how AI systems assist with complex tasks.