Computer Use Capability: Claude 3.5 Sonnet now offers a “computer use” feature in public beta, allowing developers to direct the model to interact with computers in a manner similar to human users. This includes looking at screens, moving cursors, clicking buttons, and typing text. Although this feature is still experimental and can be cumbersome and error-prone, it marks a significant step towards more agentic AI capabilities.
Tool Integration and Extended Token Outputs
The model now supports the use of Anthropic-defined computer use tools, which are integrated into the API. This enhances the model’s ability to perform complex tasks autonomously. Developers can now generate outputs up to 8,192 tokens in length using the Claude 3.5 Sonnet model, significantly increasing the model’s capacity for detailed and lengthy responses.
Prompt Caching, Efficiency, and Messages Batching API
Anthropic has also introduced prompt caching, a beta feature that allows for the caching and re-use of prompts, reducing latency by up to 80% and costs by up to 90%. The company has also introduced a Messages Batching API in beta, which allows for the asynchronous processing of large batches of queries at a reduced cost of 50%.
Workspaces and Improved API Flexibility
The Developer Console now includes Workspaces, enabling users to set custom spend or rate limits, group API keys, track usage by project, and control access with user roles. Anthropic has also loosened restrictions on the ordering of `user` and `assistant` turns in the Messages API, making it more flexible for developers to interact with the model.
