La IA de Google Gemini vence a Pokemon Blue

Google’s leading artificial intelligence project has achieved a significant accomplishment by finishing a nearly three-decade-old video game. The moment was highlighted when Sundar Pichai announced that Google Géminis 2.5 Pro successfully completed Pokémon Blue.

The gameplay feat was broadcast by a software engineer known as Joel Z, who is not employed by Google, though company leaders have supported his project. High-profile team members including Google AI Studio’s product lead shared updates as the AI progressed, sparking playful comments from Pichai about the emergence of artificial intelligence gaming within the game universe.

AI Competition in Classic Gaming

The focus on Pokémon was inspired in part by Anthropic’s recent efforts to train its Claude AI models on Pokémon Red, another vintage title within the same franchise. Claude’s journey was cited as an inspiration, though unlike Gemini, it has not yet completed its own playthrough.

Despite this apparent victory, Joel Z cautioned against using the results as a direct benchmark for comparing the performance of different language models. The methods and support systems for each AI differ, with specialized frameworks providing crucial assistance by delivering enhanced visual and contextual information from the game.

These technical harnesses allow artificial intelligence models to receive annotated screenshots, choose strategies, and even trigger specialized routines when needed. Each step forward has involved such enhancements, with both Gemini and competitive AIs making use of custom setups to navigate the complex game world.

Joel Z clarified that while some developer interventions have supported Gemini, they were made to boost general reasoning rather than provide instructions or specific solutions. No in-game walkthroughs or explicit tips were offered, aside from a single bug-related prompt that mirrored later corrections found in Pokémon Yellow.

The ongoing project continues to evolve as its creators test and improve how artificial intelligence interprets and responds to game challenges. With Gemini’s accomplishment drawing attention, industry observers are eager to see how AI technologies will further integrate with and transform vintage games in the future.