Honestly compare Claude Opus and GPT-5 on one task in 8 minutes
A blind A/B test across 5 criteria, no bias and no hate in the comments
- 1.Prepare an identical prompt-task for both models. Replace the bracketed fields with your own, after removing any confidential information from the text: client names, NDA company names, internal figures, personal data.
⚠️ Before sending to a third-party service, check this: the task text must not contain data you wouldn't post on a public blog. Use generalized phrasing: «a B2B logistics startup» instead of a company name, «sales team leads» instead of employees' names.
PromptYou are an expert copywriter. Write a 150-word email that convinces [general audience description, e.g. "small business owners"] to try [generic product category, e.g. "a project management tool"] because [generic situation, e.g. "their team wastes 5+ hours weekly on status meetings"]. Tone: friendly but professional. End with a single clear call-to-action. Do not use the words "revolutionary" or "game-changing".
What this prompt doesEnglish prompt template for the task: write a 150-word email. Replace [general audience description] with an audience type without company names, [generic product category] - with a product category, [generic situation] - with a typical problem. All three fields are anonymized descriptions, with no real names or numbers from your work. - 2.Open Claude (claude.ai), start a new chat, and paste the prompt from step 1. Copy the model's response into a separate document with the heading «Variant A». Do not label which model generated it.
A free Claude account gives access to the Sonnet model. Opus requires a Claude Pro subscription ($20/mo). If you don't have a subscription - test Sonnet, it's still a fair comparison.
Claude ↗About this tool - 3.Open ChatGPT (chatgpt.com) in incognito mode, start a new chat, and paste the same prompt word for word. Copy the response into the same document under the heading «Variant B».
The free version of ChatGPT uses GPT-4o by default. GPT-5 is available with a ChatGPT Plus subscription ($20/mo). Make sure the right model is selected in the top menu.
ChatGPT ↗About this tool - 4.Score both variants on 5 criteria from 1 to 5 points: 1) adherence to the specified length (150 words), 2) maintaining the «friendly but professional» tone, 3) absence of forbidden words («revolutionary», «game-changing»), 4) strength of the call to action, 5) natural language without clichés. Write the scores next to each variant.
Count the words in each response using any online word counter - models often exceed the limit. Search for forbidden words using Ctrl+F / Cmd+F.
- 5.Add up the scores for each variant. The variant with the higher total is the winner for your specific task. Now you can open the chats and check which model was behind A and B.
The result depends on the task type. Run the same test on a different type of prompt (code, analytics, translation), and the ranking may change. Save your prompt and scoring table for future comparisons.
- freeClaude (free account)Free access to the Sonnet model with a message limit depending on load; Opus requires a Pro subscription at $20 per month.
- freeChatGPT (free account)Free access to GPT-4o with a message limit; GPT-5 requires a ChatGPT Plus subscription at $20 per month.
Why today
AI services change fast - interfaces and free limits may differ from what's described.