Google releases Gemini 4 Hydrogen: 2.5 times faster than 3.8 Flash, overall benchmark score surpasses Argon.

Beating AI News Flash: Google releases Gemini 4 Hydrogen, succeeding the previous 3.8 Flash, focusing on everyday programming and AI agent tasks. In the latest Artificial Analysis comprehensive intelligence index, Hydrogen's default high-reasoning tier scored 51 points, surpassing the previous flagship Argon's 50 points. With the highest reasoning tier enabled, it can reach 54 points, but the time required also increases significantly.

According to Google's own testing, Hydrogen completes tasks 2.5 times faster than 3.8 Flash. On the Terminal-Bench 4.1 programming test, the score improved from the previous generation's 18.2% to 57%, though it still trails Claude Sonnet 5.5's 63.9%.

Google has also added an Orchestrator feature to Antigravity, which can automatically switch models within the same task. For example, letting Carbon handle complex architecture design and difficult problems, then handing off to Hydrogen to write code and run tests. Google's testing found that this combination can approach Carbon's standalone success rate while saving 24% of usage quota and improving task completion speed by 19%.

Hydrogen is now available to users of the Gemini app, Google Search, and AI Studio, and in Antigravity for Pro and Ultra subscribers. API pricing is $1.5 per million tokens for input and $7.5 for output, double the current discounted unit price of 3.8 Flash.