Google launched Gemini 4 Hydrogen on Wednesday, a new mid-tier model built for speed on everyday coding and agentic work, in a rollout that immediately replaces Gemini 3.8 Flash across the company's products. The announcement, made in a blog post by Koray Kavukcuoglu, SVP at Google DeepMind and Chief AI Architect, positions Hydrogen as the high-throughput counterpart to Google's frontier models, Gemini 4 Carbon and Gemini 4 Argon.
Hydrogen's debut is one of the most closely watched AI model releases of the month, and it landed just hours after Google's market-moving Gemini Decisions API news earlier this week. Developers tracking the latest AI developments will find the pricing and benchmark details below — because unlike previous Gemini 4 releases, Google made Hydrogen broadly available on day one.
A Different Rollout From Carbon and Argon
When Google released Gemini 4 Argon and Gemini 4 Carbon, access was staged: the rollout began with cybersecurity researchers, and general availability initially stayed gated behind Google's most expensive subscription tier. Ars Technica, which covered the launch, noted that Hydrogen's limits are far more relaxed, with access beginning immediately.
Hydrogen is now available across all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and Google Antigravity for paid Google AI Pro and Ultra subscribers. In the Gemini app, Ars Technica reports that free and AI Plus users remain on the older 3.5 Flash-Lite model for now, while AI Pro and Ultra subscribers get Hydrogen today.
Built for Agentic Workflows, Not Deep Reasoning
Google draws a clear division of labor in its Gemini 4 lineup. Carbon and Argon, the company's frontier models, are positioned for deep, long-horizon reasoning on the hardest problems. Hydrogen is described as "built for rapid execution across diverse agentic workflows" — writing code, running tests, verifying changes, and powering worker agents in parallel builds.
The model runs in four thinking levels: low, medium, high, and a newly introduced xhigh setting. At its default high level, Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash, and the model carries a 1 million-token input context window with support for up to 1 million output tokens.
The Benchmarks Google Is Touting
The headline claim: Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, edging past Gemini 4 Argon's 50 — and reaches 54 at the xhigh setting. For a model priced in the efficiency tier, out-scoring a frontier sibling on an independent index is a notable claim, and one that Ars Technica flagged as Hydrogen "trading blows" with Argon.
Against its direct predecessor, the gains are larger:
- Terminal-Bench 4.1: 57 percent versus 18.2 percent for Gemini 3.8 Flash — more than triple
- Fabricated claims: reduced by more than 5x, according to Google's analysis
- Partner testing: on complex graphs, SQL, and data-heavy enterprise workflows, early partners saw Hydrogen run 2x to 2.8x faster than 3.8 Flash while using 12 to 23 percent fewer tokens
Google also cited class-leading or near-top results on several agent-focused evaluations: 58.8 percent on HLE-Diamond, first place on the Vals Tax Agent Bench at 52.3 percent, a tie for first on the Vals Legal Research Bench at 55.8 percent, 53.4 percent on the new Multimodal Knowledge Agent benchmark, 71.6 percent on Agents' Last Exam v1.1, and a number-five ranking on arena.ai's live Agent Arena leaderboard.
Orchestrator: Multi-Model Sessions in Antigravity
Alongside the model, Google introduced Orchestrator in Google Antigravity, a system that dynamically switches between models such as Carbon and Hydrogen within a single session. On Terminal-Bench 4.0, Google says Orchestrator reached near-Carbon task success rates while consuming 24 percent less quota and completing work 19 percent faster on average.
The pattern Google is promoting pairs Carbon as a coordinator with Hydrogen powering worker and verifier agents across isolated workspaces — a setup aimed at teams shipping complex projects without inflating token costs.
Pricing Holds the Line at $1.50/$7.50
Gemini 4 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 — unchanged from what Gemini 3.8 Flash charged, according to Ars Technica. That pricing continuity matters for teams running high-volume agent fleets.
There is a catch elsewhere in the lineup: Ars Technica reports that Carbon and Argon currently cost $2 and $10 per million input and output tokens respectively, and that those prices will rise to $4 and $20 starting next month.
Google Search Gets Hydrogen Too
Hydrogen is also powering what Google calls the biggest update to its AI-powered search experience to date. The company says the model better handles long, multipart questions in both AI Overviews and AI Mode, that responses will include more visual elements, and that Hydrogen's fan-out search has been tuned to surface more source links. Ars Technica greeted the source-link claim with skepticism, noting Google has made similar promises before.
What It Means for Developers
For teams migrating from Gemini 3.8 Flash, Google is pitching Hydrogen as a drop-in upgrade with the same pricing, faster execution, sharper instruction fidelity, and stronger benchmark results — particularly for repository coding, debugging, deep research, and multimodal document synthesis. The model's habit of batching file-read and search calls in parallel is aimed squarely at large-codebase navigation, an area where sequential tool calls have long throttled agent throughput.
The bigger structural story is Google formalizing multi-model development: frontier reasoning models for architecture and judgment, fast execution models for the hundreds of small steps in between, and an orchestrator that moves work between them automatically. Hydrogen's day-one, all-tier availability suggests Google sees the efficiency tier — not the frontier — as where most developer demand actually lives.
Stay Ahead of AIModel launches are coming faster than ever, and the details — pricing tiers, benchmark scores, rollout gates — decide which ones matter for your stack. For breaking AI news, in-depth model analysis, and the latest AI developments, AI Buzz Wire covers every release that counts.
Read more AI news →
