Quick Overview
This video is a product walkthrough and benchmark demonstration presented by Lydia from the Claude team. It introduces the Claude Opus 5.5 model update within the Claude Code tool, detailing performance, pricing, and workflow optimizations.
Key Points
- 1.Claude Opus 5.5 is 20 percent cheaper per token than Opus 5 and provides 25 percent more usage on Pro, Max, and Team subscription limits.
- 2.Because it requires fewer tokens to complete tasks, Claude Opus 5.5 is approximately 40 percent cheaper per task overall.
- 3.In side-by-side coding benchmarks on medium effort, Opus 5.5 completed tasks 30 percent faster with fewer tool calls and more concise responses.
- 4.The default effort setting is medium, while high effort is reserved for broad codebase changes that span multiple files and dependencies.
- 5.Read-only exploration subagents can be configured to run on smaller models such as Claude Sonnet to conserve token limits.
- 6.The /claude-api prompt-audit command inspects and tunes CLAUDE.md and SKILL.md instruction files specifically for Opus 5.5.
Summary
The presentation introduces Claude Opus 5.5 and compares its performance and pricing directly against Opus 5. Opus 5.5 is faster, generates clearer and shorter summaries of its actions, and costs 20 percent less per token at 4 dollars per million input tokens and 20 dollars per million output tokens. Subscriptions on Pro, Max, and Team tiers receive 25 percent higher usage limits, and because the model reaches solutions in fewer tokens, the typical cost per task is reduced by approximately 40 percent.
To demonstrate these gains in practice, both models are run simultaneously on the same project using medium effort to fix a penny-rounding issue on discounted refunds. Opus 5.5 finishes the entire task in under one minute. Its output fits on a single screen, listing modified files, test runs, an edge case found in validation, and a single clarifying question. In contrast, Opus 5 reads unnecessary documentation and test files, executes more tool calls, and outputs a verbose response. In this comparison, Opus 5.5 runs 30 percent faster and consumes only 4 percent of the five-hour usage window compared to 6 percent used by Opus 5.
The demonstration explains the effort slider, which sets the amount of thinking the model performs before taking actions. While medium effort serves as the standard daily setting, high effort is recommended for wide-ranging changes that cross multiple files. When tasked with renaming a customer reference field across an entire codebase on medium effort, the model only updates the API handler. After switching to high effort, Opus 5.5 traces the payload through the mobile serializer, updates schema definitions, adjusts handlers, and writes new integration tests.
Additional optimizations for Claude Code workflows include configuring subagents to run on lighter models. Because exploration and read-only search agents do not require Opus-level reasoning, developers can assign them to Sonnet or Haiku directly within agent frontmatter or globally through settings.json. Finally, the built-in command /claude-api prompt-audit is introduced, which scans CLAUDE.md and SKILL.md files to eliminate redundant instructions and tune developer prompts for Opus 5.5.
Opus 5.5 Performance and Pricing Improvements
Claude Opus 5.5 introduces major speed and efficiency upgrades over Opus 5. It is priced 20 percent lower per token at 4 dollars per million input tokens and 20 dollars per million output tokens, while offering 25 percent more usage headroom across Pro, Max, and Team plans. Because Opus 5.5 solves coding tasks with fewer total tokens, the effective cost per completed task drops by roughly 40 percent.
Side-by-Side Benchmark in Claude Code
A side-by-side test fixing a financial rounding bug highlights the behavioral differences between Opus 5 and Opus 5.5. Opus 5.5 completed the task in under a minute, finishing 30 percent faster and consuming 4 percent of the five-hour session limit compared to 6 percent consumed by Opus 5. Opus 5.5 delivered a concise summary of edits, test results, edge cases, and next steps without the excessive tool calls and verbose output of its predecessor.
Managing Effort Levels and Subagent Models
Opus 5.5 defaults to medium effort, which balances speed and reasoning depth for day-to-day coding workflows. For complex tasks spanning multiple files, such as full-stack parameter renames, switching to high effort ensures comprehensive multi-file tracing and test coverage. Users can also conserve rate limits by routing exploratory read-only subagents to lighter models like Sonnet or Haiku using frontmatter or configuration settings.
Auditing Prompts for Opus 5.5
Claude Code includes a prompt audit utility triggered by running /claude-api prompt-audit. This command reads project instruction files like CLAUDE.md and SKILL.md, pruning unnecessary instructions and optimizing prompt guidelines for the updated capabilities of Opus 5.5.
The Bottom Line
The video establishes that Claude Opus 5.5 provides significant improvements in execution speed, task efficiency, and token economics over Opus 5 when used in Claude Code. It demonstrates how configuring effort levels and delegating subagents to smaller models maximizes usage limits during daily development. Unresolved questions remain regarding performance on larger production codebases outside the presented bug-fixing and refactoring scenarios.
FAQ
What is Claude Opus 5.5 and how does Claude Opus 5.5 improve upon Opus 5?
Claude Opus 5.5 is Anthropic's upgraded AI model. Compared to Opus 5, it is faster, produces clearer and more concise responses, uses fewer tokens per task, costs 20 percent less per token, and provides 25 percent more usage limit capacity on paid subscription tiers.
What does the effort setting do in Claude Code when using Claude Opus 5.5?
The effort setting dictates how much thinking time and reasoning Claude Code invests before executing actions. Medium effort serves as the standard default for routine tasks, while high effort is used for comprehensive refactoring across multiple files and services.
How much faster was Claude Opus 5.5 compared to Opus 5 in the refund bug benchmark?
Opus 5.5 completed the refund rounding bug task in under a minute, running 30 percent faster than Opus 5 while using fewer tool calls.
How can developers configure Claude Code subagents to run on Claude Sonnet instead of Opus?
Developers can set the model field to sonnet in the frontmatter of individual agent markdown files, or define the CLAUDE_CODE_SUBAGENT_MODEL environment variable in settings.json to set Sonnet as the default across all subagents.
What is the purpose of the claude-api prompt-audit command in Claude Code?
The /claude-api prompt-audit command analyzes project instruction files such as CLAUDE.md and SKILL.md, removing redundant guidance and tuning instructions to match the updated capabilities of Opus 5.5.
Worth watching for
Software developers and technical teams using Claude Code who want to understand the performance gains, pricing changes, and configuration best practices for Claude Opus 5.5.
- claude-opus-5-5
- claude-code
- ai-coding
- software-engineering
- anthropic
- developer-tools