Introduction to Claude’s New Features: A Guide to Sonnet Extended Output and Console Workbench Upgrades

This update to Claude’s new features is more geared toward developers and team collaboration: on the one hand, it enables Claude Sonnet 3.5 to produce longer outputs; on the other hand, it makes the Claude Console workbench more “evaluable and reusable.” If you’re using the Claude API for customer support, content generation, or automated workflows, these new Claude features will directly affect performance and cost.

Claude Sonnet 3.5 Extended Output: Longer answers are less likely to be cut off

In the Claude API, the maximum output token limit for Claude Sonnet 3.5 has been increased from 4096 to 8192. This new Claude feature is especially noticeable for long-form summarization, long code generation, and explanations of large tables. In the past, you might have needed to split content across multiple turns to finish it; now you can output a more complete result in one go.

Note that extended output is not “enabled by default.” When calling the Claude API, you need to add the request header: anthropic-beta: max-tokens-3-5-sonnet-2024-07-15, and set max_tokens to the range you need in order to benefit from this improvement in Claude’s new features.

Claude Console Workbench Upgrade: The prompt generator saves more time

The Claude Console workbench has added a “prompt generator.” You only need to describe the task in one sentence (for example, “categorize and handle inbound customer support requests”), and Claude will produce a more structured prompt draft. This new Claude feature is suitable for quickly turning individual experience into reusable prompt templates for the team.

In practice, it’s recommended that you fill in the inputs: goals, constraints, output format, and error-handling approach. This makes the generated prompts more stable and more aligned with Claude’s response style.

Evaluation Mode Launch: Compare prompts side by side, and score the results

The workbench’s “evaluation mode” lets you display the outputs of two or more prompts side by side and rate Claude’s outputs on a 5-point scale. This new Claude feature is ideal for A/B testing: for the same task, try different prompts to see which is more accurate, more consistent, and less verbose.

If you’re building customer support auto-replies, classification/labeling, or content moderation, it’s recommended to use “accuracy, format compliance, and deployability” as fixed scoring dimensions. Over time, this can significantly improve Claude’s controllability.

Usage and Cost Dashboard: See Claude API costs clearly

The developer console has added “Usage” and “Cost” tabs, allowing you to track usage and billing by dollar amount, token count, and API key. This new Claude feature solves an old problem: which key and which workflow is blowing through Claude’s tokens.

For teams, it’s recommended to separate different lines of business into different API keys, and then use the dashboard to cross-check and troubleshoot. This makes cost allocation for Claude clearer and also makes budgeting and setting limits easier.

Release Notes and Documentation Resources: A more complete Claude learning path

The official documentation has added more comprehensive release notes covering updates to the Claude API, Claude Console, and Claude app, making it easier for you to track changes and avoid pitfalls. Developer resources have also been updated: a newly revamped Anthropic documentation site, new training courses (Claude API basics, using Claude tools), and an expanded Claude Cookbook guide.

If you plan to connect Claude to external tools or output structured JSON, start with the “tool use” course first, then combine it with the patterns in the Cookbook—such as retrieval-augmented generation and classification—to bring Claude’s new features into your business faster.