ChatGPT 5 vs Claude Sonnet: AI Coding Skills Compared

Taha ziani — Tue, 16 Sep 2025 10:04:17 +0000

What is the future of coding? It is not just about human ingenuity but also a battle between AI giants.

Imagine two giant models, OpenAI's GPT-5 and Anthropic's Claude Sonnet, competing to build a game. One delivers careful, rule-following solutions. The other is fast and creates visually stunning designs.
Yet, neither one is perfect.

This is a glimpse into how AI shapes the tools we use to solve problems in our day-to-day life. This includes designing UIs and managing business decisions. As a developer, I rely more on AI to finish boring tasks. But which model truly gives the best results?

In this comparison, we break down the strengths, weaknesses, and surprising quirks of GPT-5 and Claude Sonnet. We look at token efficiency, pricing, and their ability to handle complex tasks like authentication. You will discover how these models compare in real-world situations. Whether you are a developer who wants precision or speed, or just curious about how AI models solve problems, this will help you understand the trade-offs. By the end, you might ask not just which model is better, but what better really means in the world of AI development.

Key Features of ChatGPT 5 and Claude Sonnet

GPT-5 is OpenAI's latest model. It is built for advanced reasoning and can adapt to many different challenges. It uses a routing system to handle tasks in the best way possible. This makes it a versatile tool for developers.

Claude Sonnet is known for its raw speed and for creating outputs that look visually refined. It is for developers who want efficiency and beautiful design.

Both models help with coding and problem-solving, but they work in dramatically different ways.

Performance in Building a Multiplayer Tic-Tac-Toe Game

We tested their coding skills by asking both models to create a multiplayer tic-tac-toe game. The results showed clear strengths and weaknesses:

GPT-5: Made a functional and robust solution. However, its interface was visually basic and needed more design work.
Claude Sonnet: Created a polished interface but sometimes added unnecessary fields, making the code more complex.

Both models had minor coding errors a reminder that human review is always required.

Impact of Rule Adherence on Model Performance

We tested how well the models followed specific rules. The models behaved very differently:

GPT-5: Was slower but more deliberate. It often gave thoughtful and detailed solutions that followed the rules closely.
Claude Sonnet: Was blazing fast but sometimes struggled with conventions. This led to small mistakes in its output.

These differences show how each model's built-in style affects its work.

Challenges in Authentication Implementation

Tasks like adding login systems were surprisingly challenging for both models. The results showed:

Both models had problems with environment variables. These issues needed manual fixes.
GPT-5: Used web searches to help. This often led to more complete and reliable solutions.
Claude Sonnet: Was fast but sometimes made functions that were incomplete or less reliable.

This shows that authentication is complex. Developers must always check the AI's output.

Token Efficiency and Cost Implications

How a model uses tokens affects its cost. This comparison showed different patterns:

GPT-5: Used more tokens because of its detailed thinking process. This often led to higher-quality outputs.
Claude Sonnet: Used fewer tokens but relied heavily on its context window. However, its higher price per token is a key factor.

This shows the trade-off between deep reasoning and token efficiency.

Pricing and Cost Analysis: Getting the Most Bang for Your Buck

Let's talk money! 💰 Cost is a massive factor when you're choosing an AI model for your projects. You want power, but you also need to keep an eye on the budget.

Here’s the latest pricing breakdown (but remember: prices can change, so always double-check the official sources!).

Side-by-Side Price Tag 🏷️

Model	Input Price (per 1M tokens)	Output Price (per 1M tokens)
GPT-5	$1.25	$10.00
Claude Sonnet	$3.00	$15.00

The Output Cost, Visualized 📊

Seeing the difference in output cost really puts things into perspective:

Cost for 1 Million Output Tokens

$20 |
$15 |            █
$10 |    █       █
 $5 |    █       █
 $0 +--------------------
        GPT-5  Claude

So, what does this mean for you?

Think of it like this:

GPT-5 is like a budget-friendly bulk store. The price per token is lower, which is fantastic for large, complex tasks where you need deep, high-quality reasoning.
Claude Sonnet might have a higher price per token, but its ability to sometimes use fewer tokens overall on shorter tasks can make its total cost competitive for quick jobs.

The bottom line? Your choice ultimately depends on your project's specific needs and budget. If you prioritize cost-efficiency on big projects, GPT-5 has the edge. For quick, token-light tasks, Claude might be your guy!

Strengths and Weaknesses of Each Model

The comparison shows clear strengths and weaknesses for each model:

🤔 GPT-5: The Deliberate Thinker

✅ Strengths: Advanced reasoning, careful problem-solving, high-quality outputs.
❌ Weaknesses: Slower, uses more tokens, sometimes makes errors.

⚡ Claude Sonnet: The Speed Artist

✅ Strengths: Extremely fast, creates visually appealing outputs, uses tokens efficiently.
❌ Weaknesses: Inconsistent with schemas, methods are less reliable, sometimes gives incomplete solutions.

Choosing the Right Model for Your Needs

So, which one is for you? The answer is: it depends.

Choose GPT-5 if your project needs accuracy, complex reasoning, and detailed outputs. It is the thinker.
Choose Claude Sonnet if your project needs speed, visual appeal, and quick iterations. It is the artist.

Both models are still improving. More testing will give us better insights into their skills, helping developers make smarter choices in the future.

Up next: What exactly is a GPT, and how does it differ from other AI models?

🏓 full article, and more are coming soon

Mangle: Google's Bold Take on Database Programming

Taha ziani — Tue, 16 Sep 2025 09:56:30 +0000

Google introduced Mangle, a new open-source programming language that extends the classic logic-based language Datalog for modern deductive databases.

Implemented as a Go package, Mangle is designed to make it easier to query and reason about data spread across multiple sources.

At its core, Mangle builds on Datalog, a declarative logic programming language with roots in database theory. While traditional Datalog is powerful for expressing complex queries, it often lacks features needed for real-world applications. Mangle bridges this gap by introducing practical extensions while keeping the accessibility and simplicity of its predecessor.

Key Features and Extensions

Mangle enhances Datalog with features that are essential for modern development, security, and data analysis:

Recursive Rules

A hallmark of Datalog, fully supported in Mangle. Recursive rules allow developers to express transitive relationships, like tracing a project's dependency tree or mapping access rights across a hierarchy.

Uniform Data Access

Mangle can treat multiple data sources as one logical database. It can pull facts from files, APIs, or databases, letting developers join information seamlessly without worrying about where it comes from.

Aggregation and Function Calls

Mangle supports aggregation functions (e.g., count, sum) and external function calls.
This makes it easier to run computations and integrate logical queries with existing code or business logic.

Optional Type Checking

Unlike plain Datalog, Mangle gives developers the option to add type safety when needed. This helps balance flexibility with reliability, especially for larger projects.

Practical Applications

Mangle is not just a research toy—it's useful for real-world problems:

Vulnerability Detection

Security teams can write rules like 'a project is vulnerable if it depends on a library with a known CVE'. Mangle can then recursively scan dependency graphs to flag issues. This makes it great for software supply chain security.

Software Dependency Analysis

Mangle is a natural fit for analysing Software Bill of Materials (SBOMs). It can help enforce versioning policies, detect deprecated libraries, or find projects relying on outdated dependencies.

Knowledge Graph Modeling

Mangle works well with knowledge graphs. By representing entities and relationships as logical facts, developers can uncover hidden connections and run advanced reasoning over complex datasets.

Implementation and Developer Accessibility

Mangle is shipped as a Go library. This means developers can embed it directly into apps and tools—no need for a separate runtime or database. The approach keeps things lightweight while putting deductive querying directly in the hands of developers.

It's worth noting that Google clearly states, “This is not an officially supported Google product.” So it's experimental, but still a powerful tool for anyone exploring deductive database programming.

In short: Mangle combines the elegance of Datalog with the practical features needed today. Whether you're a developer, SRE, or security engineer, it opens up new ways to reason about messy, distributed information—from vulnerabilities to knowledge graphs.

👉 Check out my blog for more

Forem: Taha ziani