Google releases Gemini 4 Hydrogen, outperforming larger Argon model

Google introduced Gemini 4 Hydrogen, a cost-effective model positioned below its top-tier Argon. Hydrogen replaces the previous Flash model and is 2.5 times faster than its predecessor, with hallucinations reduced by more than fivefold. In independent benchmarks on Artificial Analysis, Hydrogen ranks 7th on the highest thinking level, matching Anthropic's Claude Sonet 5.5, and surpasses Google's own Argon on the standard 'High' thinking level. This performance suggests Argon, released less than four weeks prior, is already outdated for many applications. Hydrogen is cheaper than Claude Sonet but less cost-effective than GPT 6.1 Sol. Google's top model remains Carbon, which costs double Hydrogen per task but ranks second overall, behind Anthropic's Fable 5.2. Google plans to deploy Hydrogen in its search engine to improve results and reduce hallucinations. Max Schreiner of The Decoder argues that Hydrogen is a competitive model that proves Google has successfully positioned itself in both high-end and everyday AI segments, effectively ending its previous period of stagnation.