OpenAI has recently introduced the new version of its AI, ChatGPT-5. Currently, it is trying to complete the first game of the Pokémon series faster than its competitors.
What is the challenge? Shortly after OpenAI unveiled ChatGPT-5, the AI was tasked with a challenge. It aims to complete Pokémon Red faster than previous versions or other AIs have managed to do so. A past version failed miserably.
So that every interested fan can follow the endeavor, a livestream has been set up on Twitch. There, it can be observed how the AI plays the game at a leisurely pace.
The goal is to complete the Pokémon edition and finish the league. Currently, ChatGPT-5 is on a good path. The AI has already collected all eight gym badges and is close to completing the project.
Next, the next title in the Pokémon series is coming – MeinMMO editor Jasmin Beverungen is especially looking forward to it because of a feature. Here you can see the trailer:
The AI is transparent and self-critical
What does the stream look like? On Twitch, you can closely follow how the AI approaches the game. ChatGPT-5 uses screenshots to analyze the current situation and make the appropriate decision.
All considerations the AI makes during the playthrough can be read in a separate window in the stream. This way, it is always clear what the gameplay decisions are based on. What’s particularly exciting is that ChatGPT-5 doesn’t hold back on self-criticism. If the AI makes a mistake, that too is analyzed and addressed. This allows it to adjust its strategy in the short term.
A separate window in the stream also shows which goals are next, which items, and which Pokémon are in the inventory. A dedicated minimap allows the AI to know where it is located in a simplified view, helping it remember certain locations.
Since the AI thinks through every decision carefully, completing Pokémon Red takes time. There is no quick, frantic button-mashing here. Naming a Pokémon can easily take several minutes. Accordingly, the stream has already lasted over 7 days.
In contrast to a past attempt of version ChatGPT-o3, the current AI is significantly more efficient. It only needed 3,985 interactions to obtain the first four badges in the game, while o3 needed over 10,000 actions to do so.
It will be exciting to see how long the AI still takes to reach the end. And whether future versions from OpenAI or competitors will learn to play even more efficiently. Do you follow projects like this? Feel free to write it in the comments. While playing itself, Artificial Intelligence could also play a role in the future: An AI is supposed to help you get better at your games – gives you tips while you play
