Here’s My Honest Review of GPT-5

Ever since GPT-5 dropped, the AI world hasn’t stopped speaking concerning the sheer vary of issues it may well do. Coding, writing help, picture era, even performing as an autonomous agent, it’s like having all of the issues a chatbot can do at one place. However is GPT-5 truly good? Does it actually outperform earlier OpenAI fashions? For the reason that launch, I’ve been experimenting GPT-5 with numerous prompts. I’ve listed a few of them beneath so you may attempt them too and see how the mannequin truly performs.

Earlier than we leap into the prompts, take a look at this detailed article on what GPT-5 is and the way it’s completely different from earlier OpenAI fashions.

Let’s begin going although the duties one after the other:

Objective: A shared software to trace day by day posting progress throughout platforms, have fun completions, and preserve consistency.

Customers & Roles:

1. Nitika (Social Media Supervisor) – Oversees all platforms
2. Harshit (LinkedIn Supervisor) – Posts: 4/day
3. Riya (Instagram Supervisor) – Posts: 4/day

Key Options:
✅ Each day Aim Monitoring: Visible counter for deliberate vs. accomplished posts (4/day/platform).
✅ Confetti Celebration: On the spot animated confetti when a submit is logged as “performed.”
✅ Easy Interface: Coloration-coded by platform (e.g., LinkedIn = blue, Instagram = purple).
✅ Collaboration: Notes part for every submit to share hyperlinks or feedback.

Instance Workflow:

1. Harshit logs a LinkedIn submit → counter updates → CONFETTI!
2. Dashboard reveals: *”3/4 posts performed for LinkedIn | 1/4 for Instagram”*.

Bonus: Weekly abstract report auto-generated each Friday.

Output:

Remark:

The social media tracker prototype completely executes all requested options—clear job assignments, correct submit monitoring (4/day/platform), and satisfying confetti animations upon completion. The inclusion of each day by day progress views and weekly summaries makes it sensible for workforce coordination. With its clear interface and construction (together with platform-specific coloration codes and motivational prompts), this serves as a superb developer reference. Minor enhancements like post-type categorization might strengthen V2, however the present model already delivers a strong basis.

Process 2: Create a Guess the Phrase Recreation

Create a cute and interactive UI for a “Guess the Phrase” recreation the place the participant is aware of a secret phrase and offers 3 quick clues (max 10 letters every). The AI then has 3 makes an attempt to guess the phrase. If the AI guesses accurately, it wins; in any other case, the participant wins.

Output:

Remark:

Whereas the sport delivers a enjoyable expertise with its cute UI and clean gameplay, it at the moment lacks the core characteristic the place the participant can enter the key phrase for the AI to guess. Implementing this is able to make it totally align with the unique immediate. That stated, the confetti celebration, clear design, and responsive suggestions make it an interesting prototype. With the word-input mechanic added, this might be an ideal 10/10!

Process 3: Examination Prepration

I’m making ready for an examination on Agentic AI and have lined fundamental/intermediate subjects like:

• Definition and core rules of Agentic AI
• Variations between SLMs and LLMs in agentic programs
• Function of reinforcement studying in autonomous brokers
• Moral issues in agentic AI deployment
• NVIDIA’s analysis on SLMs for agentic workflows

Create a 10-question MCQ check with:

1. 4 choices per query (single right reply)
2. Last rating report with % right
3. Detailed rationalization for any fallacious solutions, citing sources

Output:

Remark:

Wow! Killer MCQ check for Agentic AI prep! Brief however highly effective questions nail all key ideas – autonomy, instruments, ethics. On the spot suggestions explains each reply with actual examples (like how agentic AI books journeys otherwise than chatbots). Completely mimics exams with 60-second timed questions. Paste your syllabus to customise it. 10/10 for making examine enjoyable AND efficient. Greatest examination hack ever!

Process 4: Operational Duties

I needed to fill some trackers for the weekly evaluation, as a substitute of doing it manually, I requested GPT-5 to get the knowledge for me.

Give me record of all of the posts and their hyperlink posted on these channel on and afater 1st of august 2025 – https://www.instagram.com/analytics_vidhya/, https://www.linkedin.com/firm/analytics-vidhya/
Output format is a desk – Date | Post_url | platform 

Output:

Remark:

I tried to automate information assortment for weekly evaluation by asking GPT-5 to retrieve posts from Analytics Vidhya’s Instagram and LinkedIn (posted on or after August 1, 2025). The output was incomplete, whereas each platforms usually publish 4 posts per day (totaling ~25–32 posts per platform for the interval), GPT-5 returned far fewer entries.

For the reason that GPT-5 didn’t seize the total dataset precisely, I went to Manus AI and bought the duty performed!

Process 5: Reasoning and Picture Evaluation

I beforehand tried this job with OpenAI’s o3 and o4-mini and each failed at it. To know extra checkout my earlier weblog on – 6 o3 Prompts You Should Attempt As we speak. Let’s see if GPT-5 is ready to clear up this!

Present a listing of all of the particular person within the drawing together with the colour they’re drawn with.

Output:

Incorrect reply. Additionally, as this was a reasoning query the GPT-5 considering mode ought to have answered to this, however the reply was given by the traditional GPT-5 model. I chosen the considering mode manually to see if it may well reply higher. Right here’s the output:

The response stays incorrect regardless of utilizing Pondering Mode. Primarily based on this efficiency, GPT-5’s reasoning capabilities don’t seem to satisfy OpenAI’s marketed benchmarks for any such advanced question. I anticipated extra correct outcomes.

Process 6: Picture Technology

Once more, I’m attempting to match the picture era skills of GPT-5 vs GPT 4o. I beforehand tried the next immediate in my outdated article on – 4o Picture Technology is SUPER COOL.

Create a 4-image story primarily based on the next sequence:
GPT-4o believes it’s the best mannequin on the market.
GPT-4.5 arrives and surpasses GPT-4o in efficiency.
GPT-4o places in arduous work to enhance itself.
GPT-4o turns into smarter by mastering picture era.

Output:

Remark:

It’s clear that GPT-5’s picture era represents a big step backward from GPT-4o. The mannequin struggles with:

1. Textual content Rendering – Fails to precisely incorporate or show textual content inside photos
2. Picture High quality – Produces noticeably lower-resolution outputs with extra artifacts
3. Immediate Adherence – Incessantly misunderstands or ignores particular requests

For a supposedly improved mannequin, these regressions in core performance are unacceptable.

Finish Be aware

Whereas GPT-5 performs properly on coding duties, its shortcomings in reasoning, picture era, and common help (beforehand ChatGPT’s strongest promoting factors) make it a downgrade for many sensible makes use of. The enchantment of ChatGPT was its versatility as an AI assistant for on a regular basis duties, not simply coding (the place specialised instruments exist already).

Personally, I discovered the general expertise underwhelming, the mannequin didn’t ship the tangible worth I’d come to count on from earlier variations (like o3’s reasoning or GPT-4o’s picture era). The shortage of mannequin transparency (no seen indicator of which model is producing responses) solely provides to the uncertainty.

Check out some prompts your self in GPT 5 and let me know your suggestions within the remark part beneath.