Gemini 4 Hydrogen’s Everyday Help wins a LEGO game test

On 14 November, Robbert van Empel published a video comparing two Gemini 4 Hydrogen modes on the same coding task. Everyday Help and Complex Problems both received a prompt to build a playable 3D South Park LEGO game in the browser. In The Future’s benchmark, Everyday Help scored 79 out of 100 and Complex Problems scored 60. The biggest difference was in the visual category, where Everyday Help earned 31 points and Complex Problems earned 21.

Everyday Help built The Lego Globe Crisis, a snowy planet with roads, recognizable characters and LEGO studs. Players explore the globe and bring Cartman’s Cheesy Poofs, Towelie and a satellite dish back to a beacon. The radar helps them find their way, dialogue offers choices and a UFO appears at the end. There were rough edges. Part of the radar was cut off on mobile, and Kenny and a heavy brick occupied the same spot, so the interaction key kept opening his dialogue.

Complex Problems built Voxel Globe, a green sphere where players collect three boxes of Cheesy Poofs. It let players move around the globe, jump, talk to characters and drag items. The boxes were scattered without directions, and the characters were harder to recognize. After the three deliveries, the game asked players to wait for ManBearPig. No final scene followed. Both games earned nine points for execution, while Everyday Help pulled ahead on visuals, originality and interaction.

The video also compares API costs and scores from broader AI benchmarks. In its Hydrogen announcement, Google lists $1.50 per million input tokens and $7.50 per million output tokens. Those rates give developers a way to estimate costs from their own usage. The LEGO test compares two settings in the Gemini app on one task. For a browser prototype, playing through the result reveals details such as whether a quest is clear, whether the controls work on a phone and whether the game delivers the ending it promises.