Some milestones feel like a tech breakthrough. Others feel like a publicity stunt.
Google’s most advanced AI model, Gemini 2.5 Pro, has just achieved something unexpected—it beat Pokémon Blue, a game that has challenged players for nearly three decades.
The Victory That Wasn’t Just Google’s
The announcement came from Google CEO Sundar Pichai, who posted triumphantly, “What a finish! Gemini 2.5 Pro just completed Pokémon Blue!”
But here’s the catch. The Gemini Plays Pokémon livestream wasn’t actually run by Google. It was built by Joel Z, a 30-year-old software engineer who, in his own words, is “unaffiliated with Google.”
Still, Google executives have been cheering it on. Logan Kilpatrick, product lead for Google AI Studio, posted about Gemini’s steady progress, revealing that the AI had earned its fifth badge, with Pichai joking, “We are working on API, Artificial Pokémon Intelligence.”
The AI Rivalry That’s Heating Up
Why Pokémon? Back in February, Anthropic boasted that its Claude AI model was making impressive progress in Pokémon Red, showing how extended reasoning and agent training could help AI tackle unexpected challenges like classic games.
This sparked competition. There’s even a Claude Plays Pokémon Twitch channel that Joel Z cited as an inspiration for Gemini’s run.
But despite the buzz, Claude has not beaten Pokémon Red yet. Does that mean Gemini is objectively better at playing?
Not exactly.
AI Can Play—But It Can’t Play Alone
Joel Z made it clear that AI models don’t just pick up a game and start winning. Both Gemini and Claude rely on special agent harnesses, which help them process game screenshots with extra information, decide their next move, and trigger the right button commands.
And AI models still need help.
Joel Z admitted that there were developer interventions to assist Gemini, but he insisted it wasn’t cheating.
“My interventions improve Gemini’s overall decision-making and reasoning abilities,” he explained. “I don’t give specific hints—there are no walkthroughs or instructions for challenges like Mt. Moon. The only thing even close is letting Gemini know it needs to talk to a Rocket Grunt twice to get the Lift Key, which was actually a bug later fixed in Pokémon Yellow.”
The Game Is Still Changing
Gemini Plays Pokémon is still evolving. The framework continues to develop, meaning AI’s gaming skills will only get better over time.
But here’s the real question—does beating a game mean AI has mastered it? Or are humans just finding new ways to push machines toward victory?
What happens when AI models stop relying on human tweaks and start thinking ahead like real gamers?
That’s the next level.
And when AI reaches that point, the game truly changes.
