AI

Gemini 4 Hydrogen: $1.50 per million tokens and an Orchestrator

Key takeaways

  • Hydrogen is priced flat at $1.50 input and $7.50 output per million tokens across its 1 million-token window
  • Orchestrator in Antigravity cut quota use by 24% and finished 19% faster on Terminal-Bench 4.0 while nearing Carbon's success rate
  • Teams on Gemini 3.8 Flash get 2.5 times the speed and a Terminal-Bench 4.1 jump from 18.2% to 57%, though most figures come from Google's launch post

Google has released Gemini 4 Hydrogen, a model in the Gemini 4 family built for speed and efficiency in everyday coding and agentic workflows, meaning tasks where an AI model plans steps, calls tools and acts with little supervision. It is aimed at developers, enterprise teams and ordinary Gemini app users, and sits alongside Google's frontier models, Gemini 4 Carbon and Gemini 4 Argon, which excel at deep, long-horizon reasoning on the hardest problems. Google positions Hydrogen as the model built for rapid execution across diverse agentic workflows.

What it does well

The headline benchmark claim is modest but notable. On the Artificial Analysis Intelligence Index, Hydrogen scores 51 at its default high thinking level against Argon's 50, and reaches 54 at its highest xhigh setting. A model edging a frontier sibling on one index does not make it a better model overall, since the index is only one measure. It does suggest the gap between tiers is narrowing for routine work, and Google says Hydrogen often completes tasks faster than models in its class.

The more practical feature is Orchestrator in Google Antigravity. Instead of using a single model from start to finish, Orchestrator dynamically switches between models such as Carbon and Hydrogen within the same session. On Terminal-Bench 4.0 it reached near-Carbon task success rates while, on average, consuming 24% less quota and completing 19% faster. For workflows that need parallel agents, Carbon can act as the coordinator while Hydrogen powers worker and verifier agents that build and test changes simultaneously across isolated workspaces. Google says this helps teams ship complex projects faster without driving up token costs.

Hydrogen's coding behaviour is described as surgical. For issue resolution, debugging, rapid iteration and codebase Q&A it makes focused edits without rewriting unrelated files, and it natively batches file-read and search calls in parallel rather than inspecting files one by one. In early testing across complex graphs, SQL and data-heavy enterprise workflows, Google's partners saw Hydrogen run 2x to 2.8x faster while reducing token usage by 12% to 23%, compared with 3.8 Flash. It also outperforms similar-tier models on SWE-sweep, where it has to discover and repair bugs without breaking existing code.

Outside code, Hydrogen pairs native 1 million-token input with up to 1 million output tokens to synthesise dense PDFs, spreadsheets, charts, audio and recorded video into polished, cited reports and structured deliverables. Google reports 53.4% on its Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1. For deep domain reasoning it reports 58.8% on HLE-Diamond, where it leads its class, first place overall on the Vals Tax Agent Bench (52.3%) and a tie for first on the Vals Legal Research Bench (55.8%). It also leads its class on AutomationBench-AA. On multi-step tasks such as deep research, data analysis and broad codebase navigation, it ranks fifth on arena.ai's live Agent Arena benchmark.

Demos include a one-shot e-commerce storefront complete with a WebGL2 topographic shader and a live finish configurator, a Google Sheets canvas turned into an animated pixel display using Python, native =LET(...) formulas and conditional formatting, and a multi-slide Google Slides presentation built from a YouTube keynote video and a design template, followed by an autonomous audit-and-fix loop that caught overlapping text. With Gemini in Google Sheets, Google says Hydrogen can generate outputs over 30% faster than previous models in Sheets canvas, helping teams turn financial data into interactive models that adjust to forecasting scenarios. Another demo, Geminisland, lets anyone add to a shared voxel island from a single prompt.

Where it needs scrutiny

Most of the figures here come from Google's launch post. Partner results are described as "early testing", and the claim that Hydrogen "matches or outperforms some of the competing models in its class" is a selective one that does not name those rivals. Treat the benchmark list as a shortlist for your own evaluation rather than a verdict. The Artificial Analysis Intelligence Index and arena.ai's Agent Arena are external measures, but the other results are drawn from Google's launch post.

Hydrogen is also pitched at everyday work. In real-world software development, Google says teams rely on frontier models like Carbon and Argon for deep architectural judgment when designing systems or untangling subtle bugs, while still needing a fast, disciplined execution model for writing code, running tests and verifying changes. Hydrogen is built to stand on its own as a model for everyday work, or to pair with Carbon and Argon as the high-efficiency execution engine in multi-model workflows.

The four thinking levels, low, medium, high (default) and a new xhigh tier, scale across quality, latency and cost in what Google calls a smooth, predictable progression, so you only wait and pay for the reasoning depth a task requires. Teams should measure that trade-off on their own tasks, particularly at the xhigh setting that produces the 54 index score.

Orchestrator's headline result is also a ratio of savings rather than an absolute bill. "24% less quota" is an average, and the saving in any given session will depend on the workload.

How it compares

ModelRole per GoogleReported figures
Gemini 4 HydrogenRapid execution, everyday coding and agentic workflowsIndex 51 (high), 54 (xhigh); $1.50 input, $7.50 output per million tokens
Gemini 4 ArgonFrontier, deep long-horizon reasoningIndex 50
Gemini 4 CarbonFrontier, deep long-horizon reasoningOrchestrator reaches near-Carbon success rates
Gemini 3.8 FlashPrevious modelTerminal-Bench 4.1: 18.2% vs Hydrogen's 57%; Hydrogen 2.5 times faster

For context on the frontier tier, security teams weighing the two top models can read Gemini 4 Carbon vs Argon: which model security teams should pick.

Pricing and value

Hydrogen is priced flat across its entire 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens. Flat pricing means a long prompt does not trigger a higher rate partway through, which makes budgeting simpler for document-heavy jobs. UK buyers should convert from dollars at the current exchange rate and check regional availability and data-handling terms in Google Cloud and AI Studio before committing.

Output tokens are priced at five times the input rate, so output volume matters alongside the sticker price. That makes Google's claim that Hydrogen often finishes tasks while using fewer output tokens than models in its class worth testing, as is the 12% to 23% token reduction versus 3.8 Flash reported by partners. Orchestrator's quota savings are the other lever for Antigravity users. Cost per task, not cost per token, is the figure to track, a point argued in Sonnet 5.5, GPT-6.1 Sol and Argon share a price tag, not a bill.

Who should use it

Teams still running Gemini 3.8 Flash have the clearest case. Hydrogen is described as a major standalone upgrade for teams migrating from that model. Google reports tasks finishing 2.5 times faster at the default high setting, Terminal-Bench 4.1 scores more than tripling (57% against 18.2%), fabricated claims cut by more than 5x, and cleaner adaptation when a tool call fails or a search returns empty. Those are testable claims, and a migration trial on a representative sample of tickets or repositories is cheap.

Access is broad. Hydrogen is available across all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search, and to paid Google AI Pro and Ultra subscribers in Google Antigravity. Antigravity users can try Orchestrator, with Carbon coordinating and Hydrogen powering worker and verifier agents. Enterprise teams doing document, tax or legal research workflows have a reason to benchmark it against their current model, given the Vals results.

For consumers, Google says Hydrogen follows instructions more closely across multi-turn conversations, calls the right tools more often and gets users to a reliable output with less back-and-forth. It sticks closely to tone, format and specific rules, so email drafts, study guides, rewrites and documents come back ready to use, and it pulls context from uploads, Gmail, Google Drive and Connected Apps such as Airtable and Dropbox.

Teams doing deep architectural design or untangling subtle bugs should keep Carbon and Argon in the loop, since Google describes Hydrogen as able to pair with them in multi-model workflows.

Hydrogen narrows the gap between Google's tiers while keeping a flat $1.50 and $7.50 price per million tokens. The practical action for IT teams is to run a side-by-side test against Gemini 3.8 Flash and to try Orchestrator where Antigravity is already in use, measuring cost per completed task rather than headline scores.

Verdict

Worth testing now if you run Gemini 3.8 Flash or build agentic coding workflows in Antigravity;

Frequently asked questions

What is Orchestrator in Google Antigravity?

A feature that switches between models such as Carbon and Hydrogen within one session. On Terminal-Bench 4.0 it reached near-Carbon success rates using 24% less quota and finishing 19% faster.

More from Future Technology HQ