Claude API recently rounded out three things: “write longer, iterate faster, and حساب more clearly.” For those who build customer support automation, content generation, or coding assistants day to day, the biggest change isn’t the model name—it’s the detail features in the console workspace and the usage dashboard.
Claude API’s Extended Output: Sonnet 3.5 Says More in One Go
In Claude API, the maximum output token limit for Claude Sonnet 3.5 has been raised from 4096 to 8192, making it suitable for tasks like long-form summarization, solution comparisons, and explaining rules in bulk. To enable extended output, you need to add a specific beta request header in the request, rather than only changing the max_tokens parameter.
The official approach is: add "anthropic-beta": "max-tokens-3-5-sonnet-2024-07-15" to the request headers. If you find Claude API still truncates early, first check whether you missed this header and whether your max_tokens is set reasonably.
A New Prompt Generator in the Workspace: Start by Asking the Right Question
The Claude console workspace has added a “prompt generator,” and it’s straightforward: you describe the task in one sentence, such as “classify and handle incoming customer support requests,” and it will automatically produce a more complete prompt structure. For people just setting up Claude API workflows, this saves more time than hunting for templates everywhere.
A more practical way to use it is to treat the generated prompt as a baseline, then add your own field constraints, output format, and boundary conditions. This makes subsequent A/B prompt tuning in Claude API much more stable.
Evaluation Mode: Compare Prompts Side by Side—Rely Less on Gut Feel, More on Results
The workspace also provides an “evaluation mode,” which can display outputs from two or more prompts side by side and score Claude’s output on a 5-point scale. It addresses an old problem: for the same task, after tweaking just a few words, did it actually get better or worse? Often you could only go by intuition.
It’s recommended to clearly define your evaluation criteria (e.g., accuracy, formatting consistency, whether key information is cited), then use evaluation mode to run the same batch of examples. That way, iterating on Claude API prompts becomes more like engineering and less like gambling.
Usage and Cost Dashboard: Track Spending by USD, Tokens, and API Key
The developer console now includes new “Usage” and “Cost” tabs, allowing you to track Claude API consumption and billing by dollar amount, token count, and API key. For multi-environment setups (staging/production) or shared projects with multiple people, this dashboard helps quickly pinpoint “who, in what scenario, spent how much.”
If you’re optimizing costs, start here by identifying the high-consumption API keys, then go back and see whether you should shorten outputs, reduce ineffective retries, or split long tasks into more controllable segmented calls.
Release Notes and Learning Resources Filled In: No More Digging Through Messages to Find Updates
The documentation side has also been supplemented with more complete release notes, covering update logs for Claude API, the Claude console, and the Claude app—making it easier to troubleshoot “is the behavior different because the version changed?” At the same time, the official team has updated the beginner courses (Claude API basics, using Claude tools) and the expanded Cookbook, which includes hands-on paths for common capabilities such as citations, RAG, and classification.
