What is GPT 5: OpenAI's Latest Model Explained

OpenAI has released GPT 5, marking a significant upgrade from previous ChatGPT models. In this article, we’ll explore its features and test prompts, compare it with earlier models, and even build a game using GPT-5.

What is GPT-5?

GPT-5 is OpenAI’s latest flagship model, now available to everyone and replacing older versions like GPT-4.0, GPT-4 Turbo, etc. Instead of multiple separate models, GPT-5 is a unified system that combines a fast “main” model for everyday questions with a deeper reasoning model called “Thinking” that handles more complex problems. It automatically switches between these modes depending on what you need.

GPT-5 delivers detailed, step-by-step reasoning or multitool problem-solving without you having to prompt it explicitly. It also introduces multimodality as it can process text, images, audio, and even video, all within the same system. There are several versions of GPT-5 designed for different use cases:

• GPT-5 Main: The baseline, for quick and accurate answers
• GPT-5 Mini and Nano: Lightweight, cost-efficient options often used as fallbacks
• GPT-5 Thinking (limited usage for free users): For deep, thorough reasoning
• GPT-5 Pro (only available to paid subscribers): High-performance mode with extended reasoning for advanced tasks.

GPT-5 also integrates with powerful tools like Canvas for collaborative editing and deep research for citation-backed browsing, and is supposed to integrate with Gmail and Google Calendar soon. All of this comes with significantly fewer hallucinations, which is required for accurate responses. Let’s test the GPT-5 model’s capabilities with some prompts!

What does GPT-5 do?

The best way to understand what GPT-5 can really do is to test it with focused prompts, the ones that highlight reasoning, visual thinking, and how it stacks up against its predecessors. Here are some examples:

Comparing the models GPT-5 and GPT-4o

In this test, our prompt forces the model to surface measurable differences in capability and behavior. The prompt is:

Create a comparison table comparing the GPT-5 and GPT-4 models. 

Copy to clipboard
Copy to clipboard

Here is the output generated by GPT-5:

Logical reasoning with step-by-step chain of thought

This prompt asks the model to expose its internal reasoning in a structured way so you can judge coherence, missing assumptions, and error modes:

You have a small café that struggles with long queues in the morning. Suggest three quick operational changes to reduce wait times and briefly explain the reasoning behind each.

Copy to clipboard
Copy to clipboard

The response by the GPT-5 model is as follows:

GPT-5 shows clear chains of reasoning and more useful caveats than GPT-4.

Visual thinking test

This tests GPT-5’s multimodal ability to create a visual design concept, producing a hand-drawn-style wireframe you could hand to a designer:

Generate an image of a hand-drawn wireframe for a mobile app home screen that helps users track their daily water intake. Include a header with the title, a progress circle for daily water goal, quick-add buttons for logging intake, a history chart, and a motivational quote section. Style it like a sketched wireframe on notebook paper. 

Copy to clipboard
Copy to clipboard

The image created by the GPT-5 model is:

From logical reasoning to visual design, GPT-5 shows it can handle complex, creative, and analytical tasks equally easily.

Now that we’ve tested its capabilities, let’s put GPT-5 to work by building something fun, a game.

Building a riddle game with GPT-5

Now that we’ve warmed up, it’s time to get GPT-5 to build something. Let’s build a quick riddle game. The idea is: the player reads a riddle, types in their guess, and gets up to 5 chances to get it right. Each wrong answer reduces their remaining chances, and the correct answer is revealed when they run out.

Step 1: Decide the game format

Since it’s just one player, you don’t need a complex multiplayer setup. Just a basic question-and-answer flow will work. You can load a random riddle each time the player starts a new game.

If you want to create a different game, feel free to do that.

Step 2: Finalize the game flow

For our riddle game, here is a basic flow:

• Show one riddle at a time.
• The player gets exactly five chances.
• Reduce their chance count after each wrong guess.
• End the game when they guess correctly or run out of chances.
• Optionally, let them click “Play Again” for a new riddle.

Step 3: Construct a prompt to build the game

Depending on your game, you can customize your prompt. Here is a sample prompt for our game:

Build a browser-based Riddle Game where the player gets five chances to guess the answer to a displayed riddle. 
- Show the riddle and an input box for their guess. After submitting, tell them if it's correct or wrong. If wrong, decrease their remaining chances. 
- When they guess correctly, show a "You Win!" message. If they use all the chances without guessing, reveal the correct answer. 
- Add a button to start a new game with a random riddle. 
Keep the design responsive. Use HTML, CSS, and JavaScript only (no backend). 

Copy to clipboard
Copy to clipboard

Once you run this prompt, you’ll have a working game ready to test. Here is a sample code generated by the GPT-5 model:

Once the GPT-5 builds the entire code, you can select the “Run code” button to execute and view the output of this code. Here is how it looks for our prompt:

Let’s test it by guessing the output:

As we can see, it has actually created a working game based on the prompt given. From here, you can tweak the game by adding multiple riddles, hints, timers, difficulty levels, score multipliers, or even fun visuals and themes. You could also include randomization, witty wrong-answer feedback, leaderboards, or a two-player mode.

Now let’s look at what makes it so capable in the first place.

Key features of GPT-5

GPT-5 takes a big leap forward with capabilities that feel more intuitive and adaptive than ever. Here are some of its key features:

• Multimodal support: GPT-5 can understand and respond to text, images, audio, and even video cues in a single conversation. You can upload a screenshot, drop in a chart, or share an audio clip, and it will process and combine them into a useful answer.

• Advanced reasoning: This version handles complex, multi-step instructions with more accuracy and flexibility. You can pivot mid-conversation without losing context, and it will adjust its approach to match.

• Personalization and preset personalities: You can fine-tune GPT-5’s tone, depth, and style and it will maintain that personality for the entire session. From customize ChatGPT section you can pick from preset personalities or create your own, so it sounds exactly the way you want every time.

• Gmail and Google calendar integration: Plus, Pro, Team, and Enterprise users can connect GPT-5 directly to Gmail and Google Calendar. It can scan your schedule, find free time, and even draft email replies. Instead of just answering questions, it actively helps manage your day.

• Enhanced safety with safe completions: GPT-5 replaces old refusal-based safety with a more balanced approach: giving you as much safe, relevant information as possible while clearly stating limitations.

• Reasoning & verbosity controls (for developers): In the API, you can control the depth of GPT-5’s reasoning with the reasoning_effort parameter and manage response length with verbosity.

• Longer context, fewer hallucinations: GPT-5 supports a combined input and output context length of 400K tokens, and it can work with huge datasets or long-running projects without losing track. Accuracy in retrieving details from large inputs is higher, while factual hallucinations have dropped significantly.

So, how much does it cost to tap into GPT-5’s full power?

How much does GPT-5 cost?

GPT-5 is available across several plan tiers, such as Free, Plus, and Pro, with additional Team and Enterprise options for businesses.

• Free: $0/month

Includes limited access to GPT-5 and GPT-5 Mini, plus basic tools like image, voice, search, file uploads, and custom GPTs with strict usage caps.

• Plus: $20/month

Offers everything in Free, plus higher usage limits, access to voice mode with video/screensharing, deep research, multiple reasoning modes, and early access to new features.

• Pro: $200/month

Includes all Plus features and adds unrestricted usage of GPT-5 with access to GPT-5 Pro (“Thinking Pro”), OpenAI’s most advanced reasoning variant.

• Team & Enterprise: Custom pricing

Designed for organizations, it includes everything in Pro plus shared workspaces, admin controls, enterprise-grade security, connectors, and scalable usage under fair-use terms.

API/developer pricing

Developers using the OpenAI API pay per million tokens. Here are the current rates:

• GPT-5 (standard): $1.25 per million input tokens, $10 per million output tokens (cached input: $0.125)
• GPT-5 Mini: $0.25 input, $2 output (cached: $0.025)
• GPT-5 Nano: $0.05 input, $0.40 output (cached: $0.005)

Let’s explore the benchmarks that reveal GPT-5’s true capabilities.

GPT-5 benchmark performance highlights

GPT-5 delivers measurable improvements across key benchmarks that test reasoning, coding ability, and factual knowledge under strict evaluation standards. Here’s how it performs on three of the most respected AI tests.

AIME 2025 (American Invitational Mathematics Examination)

AIME is a high-level mathematics contest used to challenge the problem-solving abilities of top students. GPT-5 hit a perfect 100% score, solving every question correctly, something extremely rare even among human competitors. This shows the model’s improved ability to sustain long chains of reasoning without error, a leap beyond GPT-4o’s performance.

Source: OpenAI

GPQA Diamond (Graduate-Level Physics Questions and Answers)

GPQA Diamond is a scientific reasoning benchmark containing graduate-level physics problems designed to measure deep conceptual understanding, not just fact recall. Scoring 89.4%, GPT-5 Pro demonstrated it can parse complex questions, apply multi-step logic, and give accurate explanations. It not only beats previous OpenAI models but also edges out Google’s Gemini 2.5 Pro, signaling stronger cross-domain reasoning.

Source: OpenAI

SWE-bench Verified / Aider Polyglot (Software Engineering Tasks)

SWE-bench tests an AI’s ability to fix real bugs in large open-source repositories, while Aider Polyglot assesses multi-language coding capabilities. GPT-5 scored 74.9% on SWE-bench and 88% on Aider Polyglot, both significant improvements over o3 and GPT-4o. This means GPT-5 isn’t just good at writing code from scratch but can navigate existing codebases, identify problems, and implement correct fixes across multiple programming languages.

Source: OpenAI

These gains mean GPT-5 is better equipped for real-world use as it can tackle high-stakes math problems, explain advanced science concepts, and debug production-level code with greater accuracy than before.

Now that we’ve seen GPT-5’s benchmark scores, let’s compare it directly against its predecessors.

GPT-5 vs GPT-4o and GPT-4: What’s changed

Compared to GPT-4o and GPT-4, GPT-5 delivers sharper reasoning, faster responses, more natural multimodal interactions, and fewer hallucinations. Here’s a comparison table:

Conclusion

GPT-5 brings together speed, multimodal intelligence, sharper reasoning, and fewer errors all in a single, powerful model. We’ve walked through its feature set, seen how it performs on benchmarks, compared it with older versions, and even stretched it into games and testing prompts. It’s not just catching up, it’s pushing AI into what feels more practical and creative for everyday use.

If you’re ready to start building with GPT-5 and want hands-on practice in AI tools and prompt design, check out Codecademy’s Intro to Generative AI course.

Frequently asked questions

1. Is GPT-5 available?

Yes. GPT-5 is now the default model for ChatGPT, replacing older versions like GPT-4, GPT-4o, and Turbo.

2. Is ChatGPT-5 better?

In most cases, yes. It’s faster, more accurate, and better at complex reasoning, with expanded multimodal abilities like handling text, images, audio, and video.

3. What’s new with ChatGPT-5?

Key upgrades include integrated multimodality, deeper built-in reasoning, fewer hallucinations, smoother tool integrations (like Gmail and Calendar), and improved personalization.

4. How much does ChatGPT-5 cost?

The free tier uses GPT-5-Mini, while paid plans (Plus, Pro, Enterprise) give access to faster, more capable versions like GPT-5-Main, GPT-5-Thinking, and GPT-5-Pro. Pricing starts at $20/month for Plus.

5. Can I use GPT-5 for free?

Yes, but you’ll be limited to GPT-5-Mini, which is lighter and less capable than Pro or Thinking modes.

6. How much better is GPT-5 than GPT-4?

Benchmarks show notable improvements in reasoning, factual accuracy, and coding. In MMLU, GPQA, and SWE-Bench tests, GPT-5 outperforms GPT-4o and GPT-4 across the board.