Overview
- Google announced the Gemini 4 Hydrogen model as a new member of the Gemini 4 family focused on fast execution for coding and automated agent workflows, with staged global availability for developers, enterprises, and consumers.
- Hydrogen’s published price covers the full 1,000,000‑token context at USD 1.50 per million input tokens and USD 7.50 per million output tokens, with no segmented tiering for the large window.
- The model natively supports up to 1,000,000 tokens for both input and output and can ingest large multimodal files such as PDFs, tables, charts, audio, and video; it also offers four selectable reasoning depths (low, medium, high, xhigh) to trade cost for latency and thoroughness.
- Google introduced Orchestrator in the Antigravity platform to route tasks between models like Hydrogen and Carbon inside the same session, and reports the scheduler cut average quota use by 24% and sped completion by 19% while keeping near‑Carbon task success rates on Terminal‑Bench tests.
- Google provided vendor benchmarks showing broad gains versus Gemini 3.8 Flash — including a 2.5× default speed boost, a Terminal‑Bench 4.1 score rise from 18.2% to 57%, and large drops in hallucination and agent overreach — and said these safety and robustness figures come from company‑run test suites.