Source time Edition date
Introducing Gemini 4 Hydrogen
推出 Gemini 4 Hydrogen
Google DeepMind's Gemini 4 Hydrogen launch matters to developers and enterprises choosing a foundation model, though the sparse excerpt offers no benchmarks or pricing to judge it by.
Judgment
Teams standardizing on a Gemini tier now have to decide whether everyday coding and agent work runs on Hydrogen rather than Carbon or Argon — accepting a fast, flat-priced execution model ($1.50/$7.50 per million tokens across a 1M context) and reserving deep architectural reasoning for the frontier models.
Basis
The post is attributed to Google DeepMind's SVP and Chief AI Architect, who frames Hydrogen as bringing "cutting-edge intelligence to everyone" with "speed and efficiency for everyday coding and agentic workflows." The stated mechanism is a division of labor: Carbon and Argon handle "deep, long-horizon reasoning," Hydrogen handles "rapid execution," and Orchestrator in Google Antigravity switches between Carbon and Hydrogen inside one session — reaching near-Carbon task success on Terminal-Bench 4.0 while consuming 24% less quota and finishing 19% faster. Google puts Hydrogen at 51 on the Artificial Analysis Intelligence Index at default high thinking versus Argon's 50, and 54 at the new xhigh tier. Against Gemini 3.8 Flash it claims 2.5x faster completion at high default, Terminal-Bench 4.1 of 57% versus 18.2%, fabricated claims cut by more than 5x, and over 30% faster output in Sheets canvas. Four thinking levels (low, medium, high, xhigh) are pitched as a smooth quality-latency-cost dial.
- AA Intelligence Index, default high51
- AA Intelligence Index, xhigh54
- Terminal-Bench 4.1, Hydrogen57%
- Terminal-Bench 4.1, Gemini 3.8 Flash18.2%
Unclear The post never names the "competing models in its class" it claims to match or outperform, and gives no throughput, latency, or independent evaluation figures behind the speed claims.
Next Check Artificial Analysis' public Intelligence Index entry for Gemini 4 Hydrogen to see whether the 51 default and 54 xhigh scores appear with third-party methodology, not just Google's own comparison to Argon.