How to use Opus 5.5 and Sonnet 5.5

Summary
An official video from Anthropic's Claude channel where a presenter explains three outdated habits to avoid when using Claude Opus 5.5 and Sonnet 5.5. He demonstrates moving from micromanaged instructions to outcome- and test-driven specifications in Claude Code, dropping redundant quality prompts, and pairing Opus 5.5 for high-level reasoning with Sonnet 5.5 for execution.

What is shown

• [00:00] Release dates displayed on title cards: Claude Opus 5.5 (September 22, 2026) and Claude Sonnet 5.5 (September 28, 2026).
• [00:15] Comparison graphic showing "What I used to say" with a rigid prompt ("After the person clicks Buy, disable the payment button").
• [00:22] Graphic showing "What I say now": defining the core objective ("A user should only be charged once per order") followed by acceptance criteria ("Done when: Every payment path charges once; Every fix has a test; All existing tests pass").
• [00:30] Text graphic highlighting redundant guardrail prompts ("Make no mistakes. Double check your work.").
• [00:35] Model selection interface showing Opus 5.5, Fable 5.1, Sonnet 5.5, and Haiku 4.5.
• [00:46] Benchmark chart: Agentic coding on Terminal-Bench 4.0 comparing Sonnet 5 (10.3%) to Sonnet 5.5 (70.6%).
• [00:59] Graphic detailing target use cases for Opus 5.5 (shipping production code, system design, finance or legal analysis).
• [01:09] Graphic detailing target use cases for Sonnet 5.5 (fixing bugs, building features, documents, slides, or spreadsheets).

Claims & numbers

• The presenter notes Claude Opus 5.5 was released on September 22, 2026, and Sonnet 5.5 on September 28, 2026.
• On the Terminal-Bench 4.0 agentic coding benchmark (sourced to Anthropic), the presenter reports Sonnet 5.5 scored 70.6%, an approximately 7x increase over Sonnet 5's 10.3%.
• The presenter claims newer models self-check and avoid mistakes by default, eliminating the need for explicit quality guardrail prompts.
• Sonnet 5.5 provides faster output at lower cost than Opus 5.5 for scoped execution tasks.

Notable quotes

• [00:03] "Opus 5.5 and Sonnet 5.5 were recently released and they are amazing, but three of my old Claude habits were kind of holding these models back."
• [00:47] "Sonnet 5.5's agentic coding benchmarks are scoring nearly 7x higher than Sonnet 5's."
• [01:14] "I've been using Opus 5.5 to think critically and plan. I've been using Sonnet 5.5 to do the actual legwork."

Assessment
An official product-guidance video combining talking-head commentary with animated motion graphics and benchmark charts. The presentation is instructional rather than a live terminal capture, summarizing best-practice workflows and official benchmark metrics for Anthropic's 5.5 series models.

Described by gemini-3.8-flash on 2026-10-05 from the video's audio and frames.