Google Gemini has batted blue Pokémon (with a little help)

Google’s Model you more expensive It seems that I crossed a major major: Beating a video game of 29 years old.

Last night, Google Ceo Sundar Pickai triumphant posted on x“” That you are finished! Gemini 2.5 Pro only finished blue Pokémon! ‘

Be clear, the Gemini playing the life of pokemon was created by (in their words) “a 30 year’s software commitment declared with Google” which is going on Joel z. I am But Google executives have been issued the effort on.

For example, Logan Kilpatrick, the product lead for the Studio A Google, Posted last month which gemini was “by making a large progress in completion” and he has earned its 5th badge (the best model only has 3 and also with a different agent) ” joke“” We really have a api, artificial artificially 🙂 “

Why do Pokémon? Again in February, Antropic Highlight Highlight May your claude patterns have in “red pokémon, which the thought training” of Claude (“Pokémon Red” and “blue” are different versions of a play title first released in 1996 and linked to the long-term pokémon cannar). There is also Claude is playing twitch pokemon pokemon channel that joel z has mentioned as inspiration.

Despite his progress, Claude doesn’t look beat “Red” Pokémon. ” That means Gemini is objectively better to the game? At his twitch page, Joel Z View “, target you don’t consider this benchmark for how much one llm can play paralyon – Gemini and Claude have different tools.

And the two models you need help to play the game – is where The forefinger of agent afor Enter the patternships with playing screenshots, allowing you as you are replying to the specialized agents), and then press the button that has run with the instructions of AI.

Joela z has recognized that there were other “dev interventions to help full gems, but insists that is not cheated.

“My interventions improve the general decision of Gemini and the ability to reasoning”, says. “Do not give specific hints – there is no direct roads or instructions for the particular challenges. The only thing that comes from Gemini, that has been a lift key, which was a bug.

Plus, said, “Mailing Pukemon fucking is always actively developed, and continuous dashes continuous to evolve.”