Hey there,

The highly anticipated moment has finally arrived: OpenAI has released their new flagship model, GPT-5. With this release, OpenAI isn’t just looking to surpass their old models; they're making a clear play to outperform their top competitors like Gemini and Claude. 

So, what does that mean? Has the AI space been fundamentally changed or is this just another step on the long road to AGI? With new features like reduced hallucinations and a smart model routing system, GPT-5 is definitely making waves. 

Let’s dive into what’s new, and more importantly, how it impacts you!

What's New in GPT-5? 

As you’d expect, there have been some significant changes and improvements with OpenAI’s new model. From overall performance to more customization, we’ll now go over all the noteworthy changes. 

• Automatic Model Selection: Within ChatGPT, GPT-5 isn’t just a single model, instead it has multiple models wrapped into one. These include the following: 

  • An “efficient model” called Fast which is a fast, general purpose model

  • A “deeper reasoning model” called Thinking meant for harder questions

  • A “real-time router” called Auto that will automatically select a model based on the type of question you ask 

• Reduced Hallucinations: One of the biggest downsides of current AI models is their hallucination issues. OpenAI has made tackling this issue a priority and GPT-5 is now significantly less likely to hallucinate answers. They explained “GPT‑5’s responses are ~45% less likely to contain a factual error than GPT‑4o, and when thinking, GPT‑5’s responses are ~80% less likely to contain a factual error than OpenAI o3.” This means that you can more reliably implement AI into your workflow with greater guarantees of accurate responses.

• Improved Customization: GPT-5 is “significantly better at instruction following, and we see a corresponding improvement in its ability to follow custom instructions.” This means that GPT-5 is more easily shaped to your own needs. Additionally, OpenAI has created several unique “personalities” that can quickly change the way you interact with ChatGPT.

• Advanced Reasoning: GPT-5 has the ability to handle even more complex, multi-step tasks through GPT-5 pro. This is the highest performing model of the new GPT-5 family on many challenging benchmarks. The model made 22% fewer errors in terms of health, coding, science, and mathematics questions. 

How to Access GPT-5 

OpenAI has made the model available in 3 unique ways. These include ChatGPT, OpenAI Playground, and the API. 

• ChatGPT: GPT-5 is already available for all users in ChatGPT. This includes Plus, Team, and Enterprise users. Free users also have access, but have a much lower daily rate limit than paid users. Once free users have hit their limit, they will be transitioned to the less powerful GPT-5 mini, a faster but less powerful model.

• Playground: For developers looking to build with the new model, they can experiment with GPT-5 in OpenAI’s Playground. This testing environment will allow developers to pay as they go when using the new model.

• API: As expected, the new flagship models are also available via the API. The pricing will be the same as the Playground but will allow you to see how the new model impacts your GPT-4o powered projects. 

How Does GPT-5 Compare? 

GPT-5's performance speaks for itself. On benchmarks and platforms like LMArena (formerly LMSYS Leaderboards), the new models are showing impressive results.

The above image is from LMArena and shows that GPT-5 is a top performer in multiple areas. I chose to highlight Text and Coding here because they are two of the most popular AI applications. As you can see, GPT-5 is tied for 1st place in both categories. 

Now let's look at how GPT-5 performs in terms of standard benchmarks. I chose to highlight one of the most relevant benchmarks for this new model, hallucination rates. As I mentioned earlier, one of OpenAI’s main focuses was on reducing hallucinations across the board and they’ve definitely succeeded. 

GPT-5 in Action

Of course, benchmarks are great but the most important test lies in what people are able to create. Since its release, developers have been busy creating some apps, websites, and games using GPT-5. This Twitter Thread showcases both OpenAI created examples as well as developer created projects.