Google has introduced Gemini 4 Hydrogen, a model designed for rapid, cost‑efficient execution across developer, enterprise and consumer workflows. Within the Gemini 4 lineup, Carbon and Argon target deep, long‑horizon reasoning, while Hydrogen focuses on fast, disciplined execution for everyday coding, testing and agentic tasks.
Availability and rollout
Hydrogen is rolling out immediately across Google surfaces: Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, and for paid Google AI Pro and Ultra subscribers in Google Antigravity.
Context window, performance and pricing
Hydrogen ships with a 1,000,000‑token context window and a flat pricing model: $1.50 per million input tokens and $7.50 per million output tokens. On the Artificial Analysis Intelligence Index at the default "high" thinking level, Hydrogen scores 51 versus Argon’s 50, and it can reach 54 at the highest "xhigh" setting.
Efficiency for multi‑model workloads
Google positions Hydrogen either as a capable standalone model for day‑to‑day work or as the high‑efficiency execution engine in multi‑model workflows alongside Carbon and Argon. To facilitate that collaboration, Google introduced Orchestrator in Google Antigravity, which dynamically switches models within the same session rather than running with a single model end‑to‑end. On Terminal‑Bench 4.0, Orchestrator achieved near‑Carbon task success rates while using 24% less quota on average and completing tasks 19% faster. In parallel agent scenarios, Carbon can coordinate while Hydrogen powers worker and verifier agents to build and test changes across isolated workspaces, helping teams ship complex projects quicker without dramatically increasing token costs.
Scaling developer and enterprise workflows
Hydrogen is presented as a major upgrade for teams moving from Gemini 3.8 Flash. Compared with that predecessor, the model reportedly completes tasks 2.5× faster at its high default setting, more than triples Terminal‑Bench 4.1 scores (57% vs. 18.2%), reduces fabricated claims by over 5×, and adapts better to failed tool calls or empty search results.
Google lists specific improvements across typical workflows:
-
Iterative repository coding and debugging: Hydrogen makes focused edits, avoids rewriting unrelated files, and outperforms like‑tier models on SWE‑sweep tasks that require finding and fixing bugs without breaking existing code. For large repositories the model batches file‑read and search calls natively in parallel. Early partner tests showed 2.0×–2.8× speedups and 12%–23% token reductions compared with 3.8 Flash in complex graphs, SQL and data‑heavy enterprise workflows.
-
Open‑ended agentic execution: For multi‑step tasks such as deep research, data analysis and broad codebase navigation, Hydrogen ranks #5 on Agent Arena. It leads its class on HLE‑Diamond (58.8%), is #1 on Vals Tax Agent Bench (52.3%), ties for #1 on Vals Legal Research Bench (55.8%), and leads its class on AutomationBench‑AA.
-
Multimodal knowledge work: Hydrogen supports up to 1 million input and 1 million output tokens to synthesize PDFs, spreadsheets, charts, audio and recorded video into cited reports and structured deliverables. It scored 53.4% on the Multimodal Knowledge Agent (MKA) benchmark and 71.6% on Agents’ Last Exam v1.1 for end‑to‑end professional workflows.
-
Web development and prototyping: For tasks where speed matters—live UI prototyping, design‑to‑code iteration, interactive app building—Hydrogen produces more functional, feature‑complete layouts in fewer prompts.
Live demos and product examples
Google cites multiple demonstrations where Hydrogen generated complete artifacts from single prompts: Geminisland lets users add structures to a shared voxel island; Hydrogen plus the Gemini front end design skill in Google Antigravity built a one‑shot e‑commerce storefront with WebGL2 shader and interactive elements; Hydrogen turned Google Sheets into an animated pixel canvas using Python, native =LET(...) formulas and conditional formatting; from a YouTube keynote plus a design template, Hydrogen created a multi‑slide Google Slides presentation and ran an autonomous audit‑and‑fix loop; and Hydrogen generated interactive financial models in Google Sheets more than 30% faster than previous models.
Four thinking levels for adaptable trade‑offs
Hydrogen offers four thinking levels—low, medium, high (default) and xhigh—to trade off quality, latency and cost in a smooth progression. Across these levels, Google states Hydrogen matches or surpasses competing models in its class while often completing tasks faster and using fewer output tokens.
Consumer surfaces: Gemini app and Google Search
On consumer surfaces, Hydrogen is intended to keep multi‑turn conversations on track, follow instructions more closely, and call the right tools more frequently. In the Gemini app it aims to return near‑ready outputs for email drafts, study guides and rewrites with less back‑and‑forth. In Google Search, Hydrogen is said to better infer intent from multi‑part or typo‑filled queries, provide more visual and easy‑to‑scan responses, and surface a wider range of relevant links via improved query fan‑out.
Safety, reliability and transparency
Hydrogen incorporates the latest Gemini 4 safety and security stack and applies both built‑in training and real‑time safeguards. Google highlights two main areas of improvement:
-
Safe agentic workflows: Compared with 3.8 Flash, Hydrogen reduces agent overreach by up to 81% and under‑service by up to 79%, and cuts misaligned actions and task shortcutting in tool and coding evaluations by 12×. It is up to 2× less likely to bypass permission boundaries for impossible or blocked tasks and achieves a 96.7% honesty rate on Google’s internal agent honesty evaluation.
-
Reasoning transparency and robustness: Google reports Hydrogen preserves 99.9% chain‑of‑thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns. On the Gray Swan IPI benchmark Hydrogen is 2.3× more resilient to indirect prompt injection attacks, and robustness improvements extend across tool use, computer use and coding agents. The updated safeguards also better distinguish real biosecurity risks from legitimate medical or scientific queries; false refusals on HealthBench Pro remain under 0.4%.
Further technical and safety details are referenced in the Gemini 4 Hydrogen model card.
How to access Hydrogen
Hydrogen is available globally starting today across developer, enterprise and consumer surfaces:
- Developers: via the Gemini API in AI Studio, and in the paid tier of Android Studio; paid Google AI Pro or Ultra subscribers can also try it in Google Antigravity.
- Enterprises: deployable through Gemini for business via the API or model dropdown, Google Antigravity, and Gemini in Google Sheets for eligible business subscriptions.
- Everyone: accessible in the Gemini app model dropdown and in Google Search for all users; Gemini in Google Sheets will initially be available to Google AI Pro and Ultra subscribers.
In the Gemini app free users receive access via automatic model selection when Hydrogen is the most appropriate model for a prompt.
(Summary: Gemini 4 Hydrogen is a fast, large‑context execution model that Google positions as both a standalone productivity engine and a worker component in multi‑model orchestration, with flat pricing, four thinking levels, broad product integration and focused safety improvements.)
