Google's Gemini 4 Hydrogen Beats Larger Argon Model at a Fifth of the Task Cost

Google's Gemini 4 Hydrogen model launched at fifth place on Agent Arena with a median completion time of 5.3 minutes per task and a median cost of $1.60, running 3x faster and 4.7x cheaper per task than top-ranked Claude Fable 5.2. On Artificial Analysis evaluations, Hydrogen's default high reasoning setting matches the 51 Intelligence Index score of Google's larger Gemini 4 Argon model at 19% of the cost per task ($1.87) and a 62.5% lower token price, while its new xhigh setting reaches 54 at $3.16 per task, one point behind Claude Sonnet 5.5 at 32% of its cost. Hydrogen also recorded an 8% hallucination rate on AA-Omniscience, compared with 43% for Sonnet 5.5 and 52% for GPT-6.1 Sol.