Artificial analysis runs Gemini 4 Hydrogen using its benchmark. At the new level of thinking, the model has 54 points, one point less than Claude Sonnet 5.5 and GPT-6.1 Sol at max, and an average of 51 points: this is the level of Gemini 4 organization at five times the cost of the mission. In the AA-Omniscience study, only 8% of the questions where it was worth admitting ignorance, hydrogen answered incorrectly, Sonnet 5.5 has 43%, Sol 52%. But in the terminal it is weaker: in Terminal-Bench 4.0 it has 54% compared to 64% for Sonnet and 56% for Sol, and xHigh does not add anything here, although it almost doubles the output tokens. The index problem at xHigh is 3.16 vs 1.87 high. Therefore, to work with documents and facts, I will try hydrogen, and the encoding agent in the terminal will leave Sonnet or Sol for now. Illustrations on all levels of thinking gathered here. Channel @neuro.

Artificial analysis runs Gemini 4 Hydrogen using its benchmark. At the new level of thinking, the model has 54 points, one point less than Claude Sonnet 5.5 and GPT-6.1 Sol at max, and an average of 51 points: this is the level of Gemini 4 organization at five times the cost of the mission. In the AA-Omniscience study, only 8% of the questions where it was worth admitting ignorance, hydrogen answered incorrectly, Sonnet 5.5 has 43%, Sol 52%. But in the terminal it is weaker: in Terminal-Bench 4.0 it has 54% compared to 64% for Sonnet and 56% for Sol, and xHigh does not add anything here, although it almost doubles the output tokens. The index problem at xHigh is 3.16 vs 1.87 high. Therefore, to work with documents and facts, I will try hydrogen, and the encoding agent in the terminal will leave Sonnet or Sol for now. Illustrations on all levels of thinking gathered here. Channel @neuro.