Anthropic Just Revealed How to Prompt Opus 5

Summary
In this tutorial, presenter Paul J Lipsky reviews Anthropic's official prompting documentation for the newly released Claude Opus 5. He explains how to select appropriate models and reasoning effort settings across subscription tiers, and outlines five core prompting rules to optimize Opus 5 for knowledge work and design tasks. He then demonstrates these rules in Claude Design by generating a complete, single-page e-commerce website for a fictional brand in under three minutes.

---

What is shown

• [00:00 - 00:15] Anthropic's release page for Claude Opus 5 (dated July 24, 2026) alongside a benchmark comparison table evaluating Opus 5, Fable 5, Opus 4.8, and GPT-5.6 Sol.
• [00:16 - 00:30] Anthropic's developer documentation page titled "Prompting Claude Opus 5", highlighting behavioral differences, response length control, task scoping, and self-correction.
• [00:45 - 01:12] The Claude web interface model selector showing Fable 5, Opus 5, Sonnet 5, and Haiku 4.5, alongside effort level options (Low, Medium, High, Extra, Max).
• [01:25 - 02:36] Subscription workflow recommendations: using Sonnet by default on the $20/month Pro tier (reserving Opus 5 for complex tasks), versus defaulting to Opus 5 on the $100+/month Max plan.
• [02:40 - 03:34] Explanation of effort levels, demonstrating that Medium or Low effort is generally sufficient for standard knowledge work without excess token consumption.
• [03:41 - 08:48] Breakdown of the five prompting rules while drafting a prompt for "Northline Coffee":
  • Rule 1: Give Claude the whole job upfront instead of piecemeal steps [04:15].
  • Rule 2: Set clear scope limits so Opus 5 does not over-deliver [05:22].
  • Rule 3: Explicitly dictate the format and brevity of the final answer [06:06].
  • Rule 4: Constrain the physical length/size of the work deliverable [06:53].
  • Rule 5: Omit redundant verification instructions ("check twice") because Opus 5 auto-checks in-flight [08:00].
• [08:49 - 09:39] The finished prompt pasted into Claude Design, running Opus 5 on Medium effort with brand asset image files attached.
• [09:40 - 11:30] Generation and inspection of the complete Northline Coffee landing page rendered in Claude Design in under three minutes, verifying all six requested sections and concise bulleted deliverables.

---

Claims & numbers

• Benchmarks shown on screen [00:05]:
  • Agentic terminal coding (Frontier-Bench v2.1): Opus 5 (43.3%), Fable 5 (33.7%), Opus 4.8 (21.1%), GPT-5.6 Sol (34.4%).
  • Novel problem-solving (ARC-AGI-2): Opus 5 (30.2%), Opus 4.8 (1.5%), GPT-5.6 Sol (7.8%).
  • Agentic search (BrowseComp): Opus 5 (90.8%), Fable 5 (87.4%), Opus 4.8 (84.3%), GPT-5.6 Sol (90.4%).
  • Multidisciplinary reasoning (Humanity's Last Exam no tools): Opus 5 (56.3%), Fable 5 (56.5%), Opus 4.8 (49.8%).
  • Computer use (OSWorld 2.0 with tools): Opus 5 (70.6%), Fable 5 (66.1%), Opus 4.8 (55.7%).
  • Agentic coding (DeepSWE v1.1): Opus 5 (68.8%), Fable 5 (69.7%), Opus 4.8 (59.0%), GPT-5.6 Sol (72.7%).
• Pricing: The presenter specifies the Claude Pro plan costs $20/month and the Claude Max tier starts at $100/month [01:25, 01:44].
• Performance & Time: The presenter states that generating the complete 6-section landing page with brand assets inside Claude Design took "a little less than 3 minutes" [10:01].

---

Notable quotes

• "The effort setting mainly controls how much thinking Claude does, which affects the time and tokens it spends on the task." [02:56]
• "Opus 5 already checks and fixes its work as it goes along." [08:20]
• "Let Opus 5 use its intelligence — without making it overuse it." [11:43]

---

Assessment

This is an authentic tutorial and practical workflow review created by an independent software educator analyzing Anthropic's official documentation. The UI demonstrations in Claude Cowork and Claude Design represent real product usage, with the webpage rendering cut slightly for pacing but showcasing a working, responsive output directly adhering to the prompt constraints.

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