New Features on the Claude Developer Platform Explained: Extended Output, Evaluation Mode, and the Cost Dashboard

The focus of this Claude API update is very clear: make it easier for developers to produce longer outputs, debug prompts faster, and see usage and costs more clearly. Below, in the order of what’s “immediately useful,” I’ll break down the new features in the Claude API and the console Workbench.

Claude API Extended Output: More reliable for finishing long reports in one go

In the Claude API, the maximum output for Claude Sonnet 3.5 has been increased from 4096 tokens to 8192 tokens, making it suitable for long-form summaries, code generation, and multi-paragraph structured content. To enable extended output, you need to add the specified beta request header to your request.

Specifically, add the request header anthropic-beta: max-tokens-3-5-sonnet-2024-07-15, then set max_tokens to the range you need. For Claude API use cases that often get “cut off halfway,” this improvement is very direct.

Two practical tools in the Workbench: Prompt Generator and Evaluation Mode

The Claude console Workbench has added a “Prompt Generator.” You simply describe the task (for example, customer support ticket classification), and it will reverse-engineer a more executable prompt template, reducing the time spent building prompts from scratch. For teams looking to roll out the Claude API quickly, this step can save a lot of trial and error.

The other is “Evaluation Mode,” which lets you compare outputs from two or more prompts side by side and score the results on a 5-point scale. Using it for A/B testing is more convenient than eyeballing comparisons in a chat window, and it’s also better suited for formalizing an iterative Claude API prompt-improvement workflow.

Usage and Cost Dashboard: Break down Claude API costs clearly

The developer console has added “Usage” and “Costs” tabs, allowing you to track Claude API usage by USD amount, token count, and API key. For accounts running multiple environments (test/production) or multiple projects in parallel, this dashboard makes it faster to pinpoint “who maxed out the tokens.”

Release notes and performance items: Track Claude API changes more effectively

The documentation now includes more complete release notes, making updates to the API, the Claude console, and the Claude app easier to follow. At the same time, the Claude API also offers prompt caching (beta) to cache and reuse prompts. Official documentation says it can reduce latency by up to 80% and costs by up to 90%, making it suitable for highly repetitive system prompts or fixed knowledge-prefix scenarios.