Google Launches Gemini 4 Hydrogen as Its Fast, Cheaper Model

Google has introduced Gemini 4 Hydrogen, the third member of its Gemini 4 family and the one aimed at speed, cost and broad availability rather than maximum depth. It sits below the frontier models Gemini 4 Carbon and Gemini 4 Argon, which Google positions for long-horizon reasoning on the hardest problems. Hydrogen is meant for rapid execution: writing code, running tests, calling tools and getting through multi-step tasks without burning budget.

Pricing and reasoning levels

The pricing is simple. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens, and that rate is flat across its entire 1 million-token context window. There is no long-context surcharge to model around, which matters for teams feeding large repositories, PDFs or recorded video into a single request.

The model offers four thinking levels: low, medium, high (the default) and a new xhigh tier. Google says the progression is smooth and predictable, so developers can pay only for the reasoning depth a task needs. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at the default setting, edging past Argon's 50, and reaches 54 at xhigh. Those are Google's reported figures, and independent testing will show how they hold up in practice.

A model built to work alongside bigger ones

The most interesting part for developers is how Google frames the model's role. In real software work, teams want a heavyweight model for architecture and subtle debugging, and a cheaper, disciplined one for the bulk of edits and verification. Hydrogen is designed to be that second model, or to stand on its own.

To automate the pairing, Google is adding an Orchestrator to its Antigravity development environment. Instead of one model handling a whole session, Orchestrator switches between models such as Carbon and Hydrogen as the work changes. On Terminal-Bench 4.0, Google reports near-Carbon task success with 24% less quota consumed and 19% faster completion. For parallel-agent setups, Carbon can coordinate while Hydrogen runs worker and verifier agents in isolated workspaces.

This coordinator-and-workers pattern is increasingly common across the industry, and open-source agent frameworks have been built around it for some time. Having a vendor bake it into the tooling, with model routing handled automatically, lowers the effort of adopting it.

What it means for teams on Gemini 3.8 Flash

Google pitches Hydrogen as a direct upgrade for users of Gemini 3.8 Flash. The company claims tasks finish 2.5 times faster at the default setting, Terminal-Bench 4.1 scores rise from 18.2% to 57%, and fabricated claims drop by more than fivefold. Partners testing it on SQL, graph and data-heavy enterprise workloads reportedly saw runs 2x to 2.8x faster with 12% to 23% fewer tokens.

On the coding side, Google highlights surgical edits that avoid rewriting unrelated files, and native batching of file reads and searches in parallel when exploring large codebases. Both address common complaints about agentic coding tools: noisy diffs and slow, one-file-at-a-time exploration.

Benchmark results span legal research, tax and multimodal knowledge work, with Hydrogen ranking fifth on arena.ai's Agent Arena. As always, vendor-selected benchmarks deserve caution, and the numbers that matter most will come from your own repositories and workflows.

Safety claims and broad rollout

Google also stresses agent safety. It says Hydrogen reduces overreach in agentic tasks by up to 81%, is up to twice as unlikely to bypass permission boundaries when blocked, and is more than 2.3 times more resilient to indirect prompt injection on the Gray Swan benchmark. Prompt injection remains one of the hardest unsolved problems for agents that browse, read files and call tools, so these claims are worth testing directly.

Availability is unusually wide. Hydrogen is rolling out through Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and to paid AI Pro and Ultra subscribers in Antigravity. Search is getting upgraded handling of multi-part queries, more visual answers and a broader query fan-out that surfaces a wider range of links.

What to watch

Three things are worth tracking. First, whether independent evaluations confirm the speed and intelligence claims against competing small and mid-tier models. Second, how well Orchestrator-style routing performs outside Google's own benchmarks and whether similar controls reach the open-source agent ecosystem. Third, how the flat long-context pricing compares as rivals adjust their own rates. For developers, the practical step is simple: run Hydrogen at a few thinking levels on a real task and measure cost and latency against your current model.