Google’s Gemini AI just beat Pokémon Blue…should we be impressed?

Google’s most advanced AI model, Google Gemini 2.5 Pro, has officially cleared a decades-old hurdle – finishing Pokémon Blue, a game first released in 1996 for the GameBoy.

The win came not from Google itself, but from a livestreamed project created by Joel Z, a 30-year-old software engineer unaffiliated with the company. Still, the feat drew praise from Google leadership. “What a finish! Gemini 2.5 Pro just completed Pokémon Blue!” CEO Sundar Pichai posted on X.

Why does this matter?

Classic games like Pokémon Blue aren’t just nostalgia fodder – they’re complex environments for AI testing. Last February, Anthropic highlighted its own Claude model’s progress on Pokémon Red, pointing to the challenge as a way to demonstrate extended reasoning. Joel Z cited Claude’s efforts as inspiration, even referencing a “Claude Plays Pokémon” Twitch stream.

But Claude hasn’t beaten the game yet.

Still, don’t rush to crown a winner. As Joel Z explains, the models use different tools, known as “agent harnesses”. These setups feed the AI screenshots and layered game data, helping it make decisions and simulate button presses. Neither Claude nor Gemini is “playing” in the traditional sense. They’re interpreting structured information and issuing commands.

There were also human interventions. Joel admits to nudging Gemini occasionally, like clarifying a bugged objective involving a Team Rocket key. But he insists this isn’t cheating. “I don’t give walkthroughs or instructions,” he said. “My interventions improve Gemini’s reasoning.”

READ MORE: How to edit images using Google Gemini

Google AI Studio’s Logan Kilpatrick had been tracking progress closely. Last month, he noted Gemini had earned its fifth badge, joking that Gemini stood for “Artificial Pokémon Intelligence.”

The project isn’t over yet. According to Joel, the system continues to evolve. And while Gemini has technically “won,” the real victory may lie in what these experiments reveal about how far AI has come, and how much farther it still has to go.