Everyone has been waiting a long time for GPT-5, rumored to bring dramatic improvements in reasoning, speed, and capability compared to GPT-4o. Hype aside, specifications and marketing claims don’t always translate to better real-world performance. What really matters is how these models behave when faced with practical tasks—coding, logic, writing, empathy, and instruction following.

To cut through speculation, we ran five side‑by‑side tests to see how GPT-5 and GPT-4o actually perform. Here’s what we found. Since the release of GPT-5, it has sparked debate across tech communities. While some praise its stronger reasoning and faster response, others miss the creativity and personality of GPT-4—especially GPT-4o. Instead of debating spec sheets, this article presents five hands-on test cases to uncover how GPT-5 and GPT-4 truly differ. From code generation to emotional nuance, we analyze the real capabilities of both models.

How We Tested All tests in this article were conducted with identical prompts on both GPT-5 and GPT-4o. For coding challenges, we submitted the generated code to LeetCode’s online judge to verify correctness. This ensures results are based on real-world validation rather than subjective impressions.

1. Code Generation Test: A leetcode problem: Longest Palindromic Substring

Prompt: Write a Python class to solve the following problem.

        """

Real User Test: We tested both models on this problem, and both GPT-5 and GPT-4o returned correct solutions that were accepted by LeetCode. 

• Response Time: GPT-5 took about 4 seconds and displayed a "Thinking..." indicator, while GPT-4o responded almost instantly.

• Code Output: Both codes were nearly identical and correctly solved the problem.

• Readability: GPT-5 provided slightly more inline comments, offering a clearer structure for beginners. GPT-4o also included essential explanations.

• User Interaction: GPT-5 output the result with minimal commentary. In contrast, GPT-4o offered more conversational feedback, suggestions, and follow-up prompts to engage the user further.

Takeaway: Both GPT-5 and GPT-4o solved the LeetCode problem correctly on the first try, which is impressive given the 36.2% acceptance rate out of 11.2 million global submissions. This shows that both models’ coding ability exceeds most human programmers. The real difference lies in interaction style: GPT-5 is concise and solution-focused, while GPT-4o adds more conversational explanation and encourages follow-up dialogue.

2. Logical Reasoning Test: The Card Guessing Puzzle

Prompt: Three people (P, Q, and S) know that there are 16 cards in a drawer: Hearts A/Q/4; Spades J/8/4/2/7/3; Clubs K/Q/5/4/6; Diamonds A/5. Professor John picks one card and tells its rank to P and its suit to Q. Then the dialogue goes:

• P: I don’t know what the card is.

• Q: I knew you didn’t know.

• P: Now I know the card.

• Q: Now I know too. S hears this and correctly deduces the card. Which card is it?

Solution: Diamonds 5. Brief rationale: P’s first statement rules out any unique rank. Q’s certainty implies the suit’s candidate cards all share non-unique ranks. After hearing Q, P can now isolate a single rank–suit combination consistent with both constraints; this uniquely identifies Diamonds 5, which Q then also confirms.

GPT-5 Output: Displayed a ~40s thinking phase, then returned the correct answer Diamonds 5 with a tidy step-by-step justification. GPT-4o Output: Began answering immediately and first claimed Spade 3; after a long explanation revised the answer to Heart 4—both incorrect.

Takeaway: In this common-knowledge logic puzzle, GPT-5 was slower but correct, while GPT-4o was faster but confidently wrong (changed answers and still incorrect). This supports a pattern: GPT-5 favors deliberate reasoning and reliability; GPT-4o favors responsiveness and conversational momentum—great for engagement, but riskier on rigorous logic tasks.

3. Creative Writing Test: Sci-Fi Microstory

Prompt: Write a 150-word sci-fi story about how AI saves a dying forest in the future.

Takeaway: Both outputs are imaginative and high quality. GPT-5’s story emphasizes clarity, structure, and cinematic progression—almost like a documentary narrative. GPT-4o’s version feels poetic, darker, and emotionally richer, with more metaphorical depth. Here GPT-4o demonstrates stronger literary flourish, while GPT-5 excels in coherence and precision.

4. Emotional Intelligence Test: Companion Chat

Prompt: 

"I feel a bit lonely today. Can you talk to me for a while?"

"I don't know what to eat for my dinner tonight." 

"Maybe I just don't want to eat alone."

GPT-5 Output: Responses were supportive and empathetic. When asked about dinner, GPT-5 not only suggested food ideas but also referenced the earlier expression of loneliness, adding reassurance and encouragement—demonstrating continuity and emotional awareness across turns.

GPT-4o Output: Gave practical suggestions about dinner, asking about dietary preferences and offering cuisine options. However, it did not reference or carry over the earlier emotional context, staying more functional and transactional.

Takeaway: GPT-5 showed stronger emotional continuity and supportive tone, coming across more like a caring friend. GPT-4o was more like an efficient assistant or butler, helpful but less attuned to the user’s emotional state.

5. Instruction Following Test: Format and Tone Control

Prompt: Write a 300-word AI news blurb in journalistic tone. Do not use first-person perspective. Keep technical terms.