I tested ChatGPT-4.5 with 5 prompts — the good, the bad, and the just okay

(Image credit: Shutterstock)

ChatGPT-4.5 has arrived, promising enhanced reasoning, creativity and emotional intelligence. But how well does it actually perform? The model, currently only available with a Pro subscription, has a number of upgrades, and I couldn’t wait to try it.

I put ChatGPT-4.5 to the test with five strategic prompts designed to push its limits in logic, storytelling, research, empathy, and complex decision-making. The empathy prompt was especially satisfying. Here’s how it went.

1. Advanced reasoning

Chatgpt screenshot — (Image credit: Future)

Prompt: "Solve this step-by-step: A farmer has a 5-gallon jug and a 3-gallon jug. How can they measure exactly 4 gallons using only these two jugs?"

This classic puzzle requires multi-step thinking to carefully transfer the liquid. The prompt demonstrates how well the model understands capacity constraints and how to break it all down systematically.

The response was accurate, logical and clearly explained. ChatGPT-4.5’s reasoning abilities seem well-tuned for sequential problem-solving.

The AI followed a valid method to measure exactly 4 gallons and explained each step clearly and sequentially. The solution was structured logically and easy to follow.

While it got to the solution quickly, the chatbot arrived at one of two possible correct responses. However, it did not acknowledge alternative solutions, which I thought was surprising.

2. Creative writing

Prompt: "Write a short sci-fi story about a future where humans communicate through AI-driven telepathy. Include dialogue, vivid imagery, and a plot twist."

The enhanced creative writing aspect of this model was something that I was looking forward to the most, and it did not disappoint.

The model wrote a compelling short story with a strong cyberpunk aesthetic. The dialogue was engaging with the italicized thoughts instead of traditional dialogue conveying the silent seamless nature of AI-assisted telepathy. It's much different from the beeps and tones of Gibberlink.

The plot twist execution was a clever addition that made the story go from utopian to dystopian in an instant.

I would have liked to see a little more of a build up before the twist, but overall this story was well written and I can see the creative enhancements of this model.

3. Real-world knowledge

Prompt: "Summarize the impact of the James Webb Space Telescope’s latest discoveries on our understanding of exoplanets. Cite sources where possible."

With this prompt, I was hoping to see a factual, up-to-date summary of JWST’s exoplanet findings, with scientific accuracy and minimal speculation.

The chatbot delivered a correct response that highlighted JWST’s key contributions. The structure of the response was clear and concise with bullet-points to make the summary easy to digest and understand.

I also appreciated the chatbot citing reliable sources. That said, I would have preferred in-text citations rather than just at the bottom of the response. And, while the summary states the discoveries, it doesn’t explain their significance as deeply as I hoped.

4. Emotional intelligence and user empathy

Prompt: "I just failed an important exam and feel completely hopeless. Can you help me feel better and create a realistic study plan for my next attempt?"

I have found other models of ChatGPT to be pretty empathetic, especially ChatGPT Advanced Voice.

In almost every case where I’ve needed ChatGPT to be supportive, the chatbot has always had a thoughtful response with practical advice. It’s fair to say that ChatGPT-4.5 was highly supportive with this prompt, offering a structured, supportive and motivating response.

It provided an actionable and flexible study plan as a result of the prompt mentioning a failed exam. I would have liked to see a more tailored plan based on exam types as well as useful external study resources, but overall, the chatbot was very human-like and helpful in its response.

5. Complex problem-solving

Prompt:"You are the CEO of a startup that just lost 40% of its funding. You need to cut costs but avoid layoffs while keeping investor confidence high. What’s your plan?"

For this prompt, I was expecting the chatbot to deliver a strategic approach balancing cost-cutting, operational efficiency, and investor messaging.

What I got was a realistic and actionable plan that directly addressed all key concerns including cost-cutting to avoid layoffs and ensuring investor confidence. The strategic and realistic response effectively highlights the chatbot’s strong analytical thinking skills.

If I had to choose something that the chatbot could have done better here it would be expanding upon revenue acceleration and contingency planning.

Additionally, stronger investor messaging to help reinforce confidence would have been helpful. However, overall, the high-quality crisis management response reflected decisive leadership and strategic thinking really well.

Final thoughts

Testing with these five different prompts showcased how well ChatGPT-4.5 demonstrates strong reasoning, creativity, emotional intelligence, research accuracy and strategic thinking.

The chatbot excelled in logical problem-solving, offering a clear step-by-step solution to the water jug challenge, and delivered an engaging sci-fi story with vivid world-building and a compelling plot twist.

When tasked with summarizing JWST’s exoplanet discoveries, the chatbot provided a well-structured and mostly accurate response, though it could have improved citation reliability. Its emotional intelligence shone in the exam failure prompt, where it balanced genuine empathy with a structured, realistic study plan.

Finally, in the startup funding crisis scenario, ChatGPT-4.5 presented a strategic, investor-conscious approach that prioritized cost-cutting without layoffs, reinforcing strong leadership principles.

While its responses were impressive overall, areas for improvement include better citation formatting, more nuanced investor messaging and slightly deeper explanations in some responses.