Anthropic added a JavaScript code sandbox to its Claude AI assistant. The new tool allows users to perform complex data analysis directly in chat.
The tool acts as an integrated code environment where Claude can perform mathematical calculations, analyze data sets, and test different approaches before delivering results. All code is executed directly within the Claude.ai interface.
According to Anthropic, different teams within an organization can use the tool in different ways. Marketing teams can analyze customer interaction data to gain more in-depth insights. Sales teams can evaluate global sales performance metrics. Product managers can efficiently process user data for sprint planning sessions. Developers and finance teams can use the tool to analyze server logs or build financial dashboards.
While ChatGPT offers similar functionality, Anthropic's implementation runs code directly in the chat interface, making it more accessible to users without programming experience.
Users can enable the new analysis tool through the feature preview settings located in the bottom-left user menu. The preview also includes LaTeX support for displaying mathematical formulas and equations.
Better math and code capabilities of Claude 3.5 Sonnet are the foundation
The analysis tool builds on technical improvements made in the new Claude 3.5 Sonnet. Anthropic reports that the model has improved its programming performance on the SWE-Bench Verified Test from 33.4 to 49.0 percent, which surpasses other publicly available models. The model has also shown significant gains on the TAU-Bench test for agent-based tools.
Claude 3.5 is also Anthropic's first model with built-in computer control capabilities. The AI system can move the mouse cursor, click on items, and enter text via a virtual keyboard to perform automated browser tasks. However, these capabilities remain significantly limited compared to human users.