Lab Guide — OpenAI GPT Series

Port a Claude prompt to GPT-4o

Take the prompt comparison pattern from Lesson 03 and adapt it for the OpenAI API.

1. Set up the OpenAI client with from openai import OpenAI
2. Send a prompt to gpt-4o using client.chat.completions.create()
3. Extract the response text from response.choices[0].message.content
4. Print the token usage from response.usage
5. Compare the response structure to Anthropic’s response.content[0].text

# OpenAI response structure:
# response.choices[0].message.content  -> text
# response.usage.prompt_tokens         -> input tokens
# response.usage.completion_tokens     -> output tokens

Checkpoint: You can make a successful GPT-4o API call and extract the response text and usage stats.

Run 5 prompts on both Claude and GPT, compare outputs

Build a comparison function that sends the same prompt to both models and formats the results.

1. Define 5 test prompts covering: coding, creative writing, factual Q&A, reasoning, and summarisation
2. Write a function that sends each prompt to both GPT-4o and Claude Sonnet
3. Record: response text, token usage, and response time for each
4. Display results in a comparison table
5. Note which model gave a better response for each prompt (your judgement)

Checkpoint: A table with 5 rows showing both model responses, token counts, and timing for each prompt.

Describe an image using GPT-4o vision

Use GPT-4o’s multimodal capabilities to analyse an image.

1. Use the provided sample image URL (or any publicly accessible image)
2. Send an image analysis request using the image_url content type
3. Try different prompts: “Describe this image”, “What text is visible?”, “What emotions does this convey?”
4. Compare the detail parameter: “low” vs “high” (check token usage difference)

# Image URL content block format:
{"type": "image_url", "image_url": {"url": "https://...", "detail": "high"}}

Checkpoint: GPT-4o accurately describes the image content. You observe the token cost difference between low and high detail.