Google releases Gemini 4 Hydrogen for everyday coding and agentic workflows, with flat pricing across a 1M-token context
Synopsis
Google introduced Gemini 4 Hydrogen, a fast execution model for everyday coding and agentic workflows that complements the long-horizon reasoning models Gemini 4 Carbon and Argon. At its default high thinking level it scores 51 on the Artificial Analysis Intelligence Index versus Argon's 50, reaching 54 at the new xhigh setting; teams migrating from Gemini 3.8 Flash finish tasks 2.5 times faster, and Terminal-Bench 4.1 rises from 18.2% to 57%. Pricing is flat across its 1 million-token context at $1.50 per million input tokens and $7.50 per million output tokens.
Interpretation
Google positions Gemini 4 Hydrogen as the high-efficiency execution engine in multi-model workflows, paired with Carbon and Argon for deep architectural judgment while also standing alone for writing code, running tests and verifying changes. Teams previously relied on frontier models for both architectural design and subtle bug hunting, with no same-generation model dedicated to fast execution. At its default high thinking level, Hydrogen scores 51 on the Artificial Analysis Intelligence Index against Argon's 50, reaches 54 at xhigh, and often completes tasks faster.
Hydrogen offers four thinking levels, low, medium, high and xhigh, scaling quality, latency and cost in a smooth progression so users wait and pay only for the reasoning depth a task needs. Gemini 3.8 Flash users did not have this generation's range of reasoning-depth settings. Migrating from 3.8 Flash, Hydrogen finishes tasks 2.5 times faster at its high default setting, raises Terminal-Bench 4.1 from 18.2% to 57%, and cuts fabricated claims by more than 5x.
Google introduced Orchestrator in Google Antigravity, which dynamically switches between models such as Carbon and Hydrogen within the same session to automate multi-model collaboration. Workflows previously tended to run a single model from start to finish. On Terminal-Bench 4.0, Orchestrator reached near-Carbon task success rates while consuming 24% less quota on average and completing 19% faster.
Perspective
The report covers Google's published release information and self-reported evaluations: the intelligence index, Terminal-Bench, Agent Arena, HLE-Diamond, Vals and MKA scores all come from Google's materials, and the intended readers are developers and enterprises weighing model selection and agentic workflows.
Several performance and safety figures come from Google's internal evaluations or early partner testing, such as partners observing 2x to 2.8x speedups and 12% to 23% lower token usage on complex graphs, SQL and data-heavy enterprise workflows; these results have not been independently replicated, and real-world behavior may vary with task and deployment.