How Anthropic Engineers Actually Use Claude Opus 5.5

Summary
Duncan Rogoff reviews an Anthropic engineering guide titled "Getting the most out of Opus 5.5 in Claude and Claude Code," authored by Addy Osmani. The video walks through key operational changes, prompting practices, and workflow adjustments recommended for using Claude Opus 5.5 effectively in coding and agentic tasks.

What is shown

• [00:08] The official announcement page and benchmark comparison table for Claude Opus 5.5 versus Fable 5.1, Opus 5, GPT-6 Astra, and GPT-5.6 Sol across evaluations like Terminal-Bench 4.0 and CursorBench 4.0.
• [00:34] The playbook article "Getting the most out of Opus 5.5 in Claude and Claude Code" on claude.dev/blog.
• [00:51] First core guideline: defining what "done" means in a single prompt and letting the model execute autonomously.
• [01:30] Recommendation to delete "think carefully" or "think step by step" prompt instructions since Opus 5.5 has integrated thinking before replies.
• [02:31] Concrete prompt example showing migration instructions with explicit completion conditions and stopping triggers.
• [03:32] Demonstrating mid-run user input in Claude Code to steer execution without restarting context or waiting for a complete run to end.
• [04:08] Design prompting techniques: enumerating specific negative style constraints (e.g., avoiding cream/off-white backgrounds, italic accents, pill-shaped buttons).
• [04:54] Configuring steering rules inside CLAUDE.md to define when Claude should autonomously continue versus stopping to request confirmation.
• [06:02] Splitting large code audits and migrations across subagents in parallel.
• [06:23] Using an external checklist file (TASKS.md) to retain progress tracking across context compaction and summarization during extended sessions.

Claims & numbers

• The presenter states that Claude Opus 5.5 was released on September 22, 2026.
• The presenter states that Claude Opus 5.5 scores 66.4% on Terminal-Bench 4.0, 54.4% on FrontierCode v1.1, and 57.8% on CursorBench 4.0, outperforming Fable 5.1, GPT-6 Astra, and GPT-5.6 Sol.
• The on-screen pricing table displays Opus 5.5 pricing as $5 per million input tokens, $25 per million output tokens, $0.20 cache read, and $5 cache write, with claims that it costs 40% less to run than Opus 5.
• The presenter claims fast mode for Opus 5.5 is available in Claude Code and Claude Platform with up to 2.5x speed, costing $8 per million input tokens and $40 per million output tokens.
• The presenter states that removing "think carefully" instructions in testing resulted in replies starting sooner with no measurable loss in response quality.

Notable quotes

• [00:43] "It works for longer on its own, it tells you plainly what it did, which is super nice, and it thinks before every reply."
• [03:13] "In our testing in a chat product, removing a 'think carefully' line made replies start sooner, with no clear drop in quality."
• [04:21] "Don't just give it direction, tell it exactly what you don't want."

Assessment
This is a walkthrough and commentary video analyzing an official Anthropic blog post and documentation release. The presenter shows authentic screens of the published guide and benchmarks, summarizing official advice without performing live coding demonstrations directly on camera.

Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.