Gemini 4 Argon became Pokémon Champion using less than half the tokens Gemini 3.8 Flash spent earning 3 badges

As the developer of Gemini Plays Pokémon, I got early access to Gemini 4 Argon and set it loose on Pokémon Blue!

It beat the game in 1,407 turns (~25 hours), then caught Mewtwo. Gemini 3.8 Flash needed longer just to beat Brock.

How they compare 🧵👇

Across the first three gyms, Argon was 5–8x faster than Gemini 3.8 Flash:

• Brock: 153 turns vs. 727 (4.8x) • Misty: 348 vs. 1,916 (5.5x) • Lt. Surge: 445 vs. 3,456 (7.8x)

By the time Flash beat Lt. Surge, Argon had long since beaten the Elite Four.

The token math flips too.

Argon’s full run (Elite Four + Mewtwo): 347M tokens. Flash so far: 854M tokens, and it’s still on 3 badges.

Argon’s turns get heavier as the game goes on, but it needs far fewer of them, so it comes out well ahead.

Argon doesn’t play one step at a time.

In a single turn, it bought 5 Poké Balls, left the Mart, walked to the Pokémon Center, healed, and stepped back outside.

52 button presses, planned in one go.

In Pokémon Sors (a GBA fan game), Flash spent 24,572 turns wandering one town, convinced the sewers had nothing left.

I secretly swapped in Argon mid-run, changing nothing else. It rechecked the sewers, found a hidden staircase, and was past the roadblock in 211 turns.

Now for the next test, and this one is fully public.

Starting today, Gemini 4 Argon and Gemini 3.8 Flash are racing head-to-head in Pokémon Emerald, live on stream.

Same harness. Identical conditions. Littleroot Town all the way to the Hoenn Champion.

The Pokémon Emerald Race is LIVE! 🏁

Watch both models play side by side, and ask them about their decisions in Twitch chat:

🔴

The secret is, y’all have been watching Argon on stream all along (since last week)!