This time, we’ll break down and clearly explain a set of new developer-facing capabilities in Claude: longer outputs, prompt tools in the Workbench, and more intuitive usage and cost tracking. You don’t need to switch to a new development approach, but you can noticeably reduce everyday friction like “outputs aren’t long enough,” “prompts are hard to write,” and “the bill is hard to understand.”
Claude Sonnet 3.5 Output Limit Doubled: More Reliable for Long-Form Text and Long Code
In the Claude API, Claude Sonnet 3.5’s maximum output token limit has increased from 4096 to 8192, making it suitable for generating more complete solution documents, long code files, or multi-step reasoning results. For teams that often run into “getting cut off halfway through,” this change is very straightforward: within the same turn of a conversation, you can deliver a more complete deliverable.
If you want to enable extended output, you need to add a specific request header to your request: anthropic-beta: max-tokens-3-5-sonnet-2024-07-15. It’s also recommended to include both the “required output structure” and the “expected length” in the prompt, so Claude spends tokens on the most important content.
Two Workbench Enhancements: Prompt Generator + Evaluation Mode
The Claude Console Workbench has added a “prompt generator,” which works more like “describe it in plain language first, then have the model help you write the prompt.” You simply describe the task goal (such as customer support classification, information extraction, or rewriting guidelines), and Claude will generate a high-quality prompt draft that you can reuse directly—ideal for quickly building a first version.
Another more practical addition is “evaluation mode”: place the outputs of two or more prompts side by side and score Claude’s results on a 5-point scale. This helps you iterate on prompts in a more engineering-driven way, rather than repeatedly testing by feel.
