If you often use Claude for long-form generation, code explanations, or batch-processing tasks, this developer-side update is very practical: longer outputs, a more usable workbench, and more transparent billing. Below, I’ll break it down by feature to clearly explain how to use it and what scenarios it fits.
Claude Sonnet 3.5 Extended Output: From 4096 to 8192 tokens
In the API, Claude Sonnet 3.5 doubles the maximum output token limit from 4096 to 8192, making it better suited for generating a complete solution in one go, long reports, or longer code files. To enable extended output, you need to add a beta request header to the request, rather than only changing the max_tokens parameter.
The official format is: "anthropic-beta": "max-tokens-3-5-sonnet-2024-07-15". In practice, it’s recommended to also keep prompt length and output targets under control; otherwise, Claude is more likely to “think more” but drift away from the intended structure.
Claude Console Workbench: The Prompt Generator Saves Communication Costs
The newly added “prompt generator” in the Claude console workbench is suitable for teammates who aren’t good at writing prompts: you simply describe the task in natural language (for example, “classify and handle incoming customer support requests”), and Claude will produce a more standardized prompt template.
I’d recommend treating it more as a “prompt drafting tool”: first have Claude generate the framework, then fill in your field definitions, output format, and boundary conditions—this can significantly reduce trial-and-error iterations.
Evaluation Mode: Side-by-Side Comparison of Output Quality Across Different Prompts
Another useful feature is “evaluation mode,” which can display the results of two or more prompts side by side and score Claude’s outputs on a 5-point scale. For scenarios that require stable outputs (content moderation, classification, information extraction), this is more reliable than picking prompts by gut feel.
