Google’s Gemini has beaten Pokémon Blue (with a little help)

Google’s most expensive AI model seems to have crossed a large milestone: beating a 29-year-old video game.

Last night, Google CEO Sundar Pichai triumphantly posted on X“What a finish! Gemini 2.5 Pro has just completed Pokémon Blue!”

To be clear, the Gemini plays Pokemon Livestream was made by (in his own words) “a 30 -year software -engineer not affiliated with Google” that passes by Joel Z. But Google leaders have fueled the efforts.

Logan Kilpatrick, the product leader for Google AI Studio, for example, Posted last month Die Gemini was “making great progress when completing Pokémon” and had “earned his 5th badge (the next best model has only 3 so far, although with another agent harness),” leading pichai to joke“We are working on API, artificial Pokémon -Intelligence :)”

Why Pokémon? Back in February, Anthropic emphasized progress These are Claude AI models made in “Pokémon Red”, writing that Claude’s “Extended Thinking and Agent Training” The “A big boost” gives on “more unexpected” tasks, such as playing a classic game. (“Pokémon Red” and “Blue” are different versions of A GameBoy title First released in 1996 and connected to the long-term Pokémon franchise). There is even A Claude plays Pokemon Twitch -Canaal That Joel Z called as inspiration.

Despite the progress, Claude does not seem to have defeated “Pokémon Red”. Does that mean that Gemini is objectively better in the game? On his Twitch page, Joel Z insisted on viewers: “Please do not consider this as a benchmark for how well an LLM Pokemon can play. You can’t really make direct comparisons – Gemini and Claude have different tools and receive different information.”

And both AI models need help to play the game – that’s true Use the aforementioned agent Come in and offer the models game screenshots covered with additional information, so that the model can decide how to respond (where specialized agents can be invoked) and then press the button that corresponds to the instruction of the AI.

Joel recognized that there were other “DEV interventions” to help Gemini complete the game, but it insisted that it does not false.

“My interventions improve Gemini’s general decision-making and reasoning options,” he says. “I don’t give specific hints – there are no passage or direct instructions for certain challenges such as Mt. Moon.

Moreover, he said: “Gemini plays Pokémon is still being actively developed and the framework continues to evolve.”