Google has officially unveiled Gemini 4 Hydrogen, expanding its next-generation artificial intelligence family alongside Gemini 4 Carbon and Argon. Built to deliver high-speed, cost-effective intelligence for everyday coding, professional document drafting, and agentic workflows, Hydrogen is rolling out globally today across web, mobile, Google AI Studio, the Gemini API, and Google Antigravity.
Key Highlights and Benchmark Supremacy
- High-Fidelity Agentic Execution: Hydrogen is optimized for tight instruction following, strict formatting adherence, and automated multi-step tool calling across Google Workspace apps like Gmail and Google Drive, as well as third-party services like Airtable and Dropbox.
- Terminal-Bench and SWE-sweep Breakthroughs: The model achieves a 57.0% cheating-adjusted score on Terminal-Bench 4.1, more than tripling Gemini 3.8 Flash (18.2%), while reaching 13.5% on SWE-sweep to outpace Claude Sonnet 5.5 (3.1%) and GPT-6.1 Terra (5.8%).
- Beating Argon on CursorBench 4.0: In coding evaluations, Gemini 4 Hydrogen High scores 54.3% on CursorBench 4.0, edging out Gemini 4 Argon High (52.9%) while consuming fewer execution steps and lower token costs ($5.59 per task).
- Artificial Analysis Intelligence Index: Hydrogen scores 51 on the Artificial Analysis Index, climbing to 54 at its highest reasoning setting, surpassing Argon (50).
- Autonomous Commerce and Reasoning: Generates a top-tier $14,549 bank balance on Vending-Bench 2, alongside a 71.6% score on Agents’ Last Exam v1.1 and 58.8% on HLE-Diamond.
- Antigravity Paid Subscription Gate: While accessible across all Google AI Studio tiers, Antigravity access is strictly restricted to paid Google AI Pro, Ultra, and enterprise subscribers, leaving promotional plans without access.
Google Gemini 4 Hydrogen: Benchmark Performance and Deployment Matrix
| Evaluation Metric / Category | Gemini 4 Hydrogen | Claude Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|
| SWE-sweep (Agentic Coding) | 13.5% | 3.1% | 5.8% |
| Terminal-Bench 4.1 (Cheating-Adjusted) | 57.0% | 63.9% | Unavailable |
| CursorBench 4.0 (Coding Score) | 54.3% (Hydrogen High) | 55.5% (Sonnet 5.5 Max) | N/A |
| Artificial Analysis Intelligence Index | 51 (Base) / 54 (High) | 52 | 51 |
| Agents’ Last Exam v1.1 (Knowledge Work) | 71.6% | 60.4% | 69.8% |
| AutomationBench-AA | 76.1% | 71.8% | 60.2% |
| Vending-Bench 2 (Autonomous Balance) | $14,549 | $11,165 | $9,107 |
| OSWorld-2.1 (Computer Use, Offline) | 77.6% | 83.9% | 74.2% |
| MKA (Multimodal Understanding) | 53.4% | 52.0% | 43.8% |
| Chartography (Visual Analysis) | 66.6% | 61.6% | 63.2% |
| HLE-Diamond (Advanced Reasoning) | 58.8% | 53.3% | 47.3% |
| Antigravity Availability | Paid Pro, Ultra, Enterprise | Via API / Workbench | Via Platform |
| Developer Tiers | Google AI Studio (All), Gemini API | Anthropic Console | OpenAI Platform |
Deployment Channels and the Jio Pro Limitation
Google is deploying Gemini 4 Hydrogen across its full consumer and developer ecosystem. For everyday users on the web, Android, and iOS, the model is integrated into the core Gemini app and Google Search AI Mode via dynamic automatic model routing. Developers can immediately access Hydrogen across all tiers in Google AI Studio and through the standard Gemini API for commercial integration.
However, the rollout in Google Antigravity features a specific commercial distinction. Google has designated Hydrogen as an entitlement for paid Google AI Pro, Google AI Ultra, and enterprise workspace accounts. In India, where millions of subscribers activated promotional Google AI Pro benefits bundled with Reliance Jio telecom plans, users report seeing Hydrogen marked as unavailable with the notice: “Gemini 4 Hydrogen is currently available to paid Pro and Ultra subscribers.” Because the promotional Jio bundle is classified as a sponsored partner entitlement rather than a direct paid recurring subscription, Antigravity currently restricts model access. Indian developers relying on bundled plans can still evaluate Hydrogen freely via Google AI Studio.
This launch follows Google’s rapid expansion of dedicated foundation models. Earlier this week, Google launched Gemini 4 Carbon with Antigravity Swarm support for heavy multi-agent software engineering, while also deploying Gemini Decision 1 for fast AI agent routing. Meanwhile, competition in frontier development tools continues to intensify as Anthropic unveiled Claude Fable 5.2 and Mythos 5.2 with aggressive developer pricing.
PFD Context
Gemini 4 Hydrogen represents a strategic realignment for Google’s agentic ecosystem. Rather than competing purely on raw parameter scale against frontier reasoning engines, Hydrogen focuses on token economics, tool precision, and latency. By outpacing Argon on CursorBench 4.0 and crushing Gemini 3.8 Flash on Terminal-Bench, Google provides developers with an ideal daily driver model for interactive IDE sessions. The strict subscription tiering in Antigravity underscores how foundational AI tooling is increasingly tied to direct subscription monetization, separating commercial paying users from telecom bundle recipients.
