Claude API’s New July Features: Sonnet 3.5 Launch and Guide to Doubling the Output Limit

What problem does this Claude API update solve?

In the July 2024 Claude API update, the focus centers on Claude Sonnet 3.5: stronger overall capability, more stable speed and cost performance, and a more complete developer experience. For teams that need long-form content generation, batch processing, and cost visibility, this upgrade is the kind you can “use right away.”

If you previously often got stuck in the Claude API with output length limits, unstable prompt quality, or difficulty tracking usage, this update basically provides the official answers to all of those.

Claude Sonnet 3.5 Launch: Stronger, while keeping mid-tier model cost

Claude API now provides Claude Sonnet 3.5, which Anthropic emphasized in its announcement as the “latest version,” noting that it outperforms competitors as well as Claude Opus 3 on multiple evaluations, while still maintaining the speed and cost positioning of a mid-tier model. For everyday applications (customer support categorization, summarization, writing, code assistance), switching to Claude Sonnet 3.5 often directly reduces rework.

For real-world rollout, it’s recommended to first run an A/B comparison in the Claude API using the same set of prompts: check output consistency, factuality, and adherence to formatting before deciding whether to switch over fully.

Max output increased from 4096 to 8192: How to enable it in the Claude API

In the Claude API, the maximum output tokens for Claude Sonnet 3.5 have doubled from 4096 to 8192, but it must be explicitly enabled. The official method is to include this header in the request: anthropic-beta: max-tokens-3-5-sonnet-2024-07-15, and then set max_tokens to the value you need.

This is especially useful for long-form summarization, report generation, and long code completion: you can reduce the number of “continue writing” rounds, keep overall latency more controllable, and reduce the chance of drifting off-topic across multi-turn conversations.

Workbench Prompt Generator: Hand the “prompt writing” step to a tool

In the enhanced Workbench in the Claude Console, a new “prompt generator” has been added. You simply describe the task (for example, “categorize incoming customer support requests”), and it will generate a more structured, more reusable prompt—useful for turning individual experience into team standards.

It’s recommended to finalize prompts in Workbench before going live on the Claude API: lock down input/output formats and add boundary conditions, then copy into production. This can noticeably reduce the cost of prompt iteration.

Cost and documentation updates: Claude API development is easier to manage and easier to learn

This update also refreshes the usage and cost dashboard, making it easier to review Claude API consumption by time and volume, and to clearly account for load testing and staged rollouts. Meanwhile, Anthropic also mentioned a documentation revamp, new educational courses, and an expanded Claude Cookbook—overall leaning more toward “ready-to-use, follow-along” developer resources.

If you’re preparing to integrate the Claude API into your business workflow, it’s recommended to start with the patterns in the new documentation and Cookbook, then adjust model selection and max_tokens strategy based on dashboard data to avoid maxing out costs from the outset.