Gemini 4 Hydrogen: Fast Model, Four Thinking Levels
Google's Gemini 4 Hydrogen targets everyday coding and agents at $1.50/$7.50 per million tokens, with four thinking levels and wide rollout.
Google's Gemini 4 Hydrogen targets everyday coding and agents at $1.50/$7.50 per million tokens, with four thinking levels and wide rollout.
Introduction
On November 12, 2026, Google introduced Gemini 4 Hydrogen, which it describes as built "for rapid execution across diverse agentic workflows." It joins the frontier models Gemini 4 Carbon and Gemini 4 Argon, which Google says excel at deep, long-horizon reasoning, as the fast tier of the Gemini 4 family. Google's headline claims are a score of 51 on the Artificial Analysis Intelligence Index at its default high thinking level, against 50 for Argon, and 54 at the new xhigh setting. Pricing is $1.50 per million input tokens and $7.50 per million output tokens across the whole 1 million-token context window. Ars Technica reports that Hydrogen replaces Gemini 3.8 Flash across Google's products. This review covers Hydrogen only. Carbon, a gated cyber-defense release, was covered in a separate Evermx article.
Feature Overview
Four thinking levels. Hydrogen offers low, medium, high (the default) and a new xhigh tier. Google says they scale in a smooth, predictable progression, so users wait and pay only for the reasoning depth a task requires.
Gains over Gemini 3.8 Flash. Google positions Hydrogen as a major upgrade for teams migrating from 3.8 Flash. All figures below are Google-reported.
| Measure | Gemini 3.8 Flash | Gemini 4 Hydrogen |
|---|---|---|
| Terminal-Bench 4.1 | 18.2% | 57% |
| Speed at default high thinking | Baseline | 2.5x faster |
| Fabricated claims | Baseline | Cut by more than 5x |
Google also says partners saw Hydrogen run 2x to 2.8x faster while using 12% to 23% fewer tokens than 3.8 Flash in early testing across complex graphs, SQL and data-heavy enterprise workflows.
Long context and long output. Hydrogen takes up to 1 million tokens of input and can produce up to 1 million output tokens, aimed at dense PDFs, spreadsheets, charts, audio and recorded video turned into cited reports.
Agentic and domain benchmarks. Google reports that Hydrogen ranks #5 on Agent Arena, arena.ai's live agent benchmark, and scores 58.8% on HLE-Diamond, which Google says leads its class. It also reports 52.3% on Vals Tax Agent Bench (#1 overall), 55.8% on Vals Legal Research Bench (tied for #1), 53.4% on the new Multimodal Knowledge Agent benchmark and 71.6% on Agents' Last Exam v1.1.
Orchestrator in Google Antigravity. Antigravity gains a model option called Orchestrator that switches between Carbon and Hydrogen within one session. As an example, Google describes Carbon starting in a read-only planning phase, Hydrogen then writing edits and running tests, and a hand-back to Carbon if Hydrogen hits friction. Google says Orchestrator reached near-Carbon task success while consuming 24% less quota and finishing 19% faster on average. Google's launch post cites Terminal-Bench 4.0 for this result, while the Antigravity post cites Terminal-Bench 4.1.
Search and the Gemini app. Google is bringing Hydrogen to Search for all users globally, calling it the biggest model upgrade to Search yet. It says Search will better understand multi-part questions and typos, return more visual answers, and use an upgraded query fan-out to link to a broader range of sites.
Safety figures. Google reports the following from its own evaluations: overreach in agentic tasks down by up to 81%, under-service down by up to 79%, misaligned actions and task shortcutting cut 12x, and up to 2x less likely to bypass permission boundaries on impossible or blocked tasks. It cites a 94% score on its internal agent honesty evaluation, 99.9% chain-of-thought legibility across more than 200,000 audited turns, 2.3x more resilience on the Gray Swan indirect prompt injection benchmark, and false refusals under 0.4% on HealthBench Pro.
Usability Analysis
For developers, the practical choice is the thinking level. High is the default; low and medium trade depth for latency and cost, and xhigh is the setting behind the 54 index score. Hydrogen is available through the Gemini API in Google AI Studio and in Android Studio, and in Gemini for business. Paid Google AI Pro and Ultra subscribers can use it in Antigravity, and Gemini in Google Sheets starts with AI Pro and Ultra.
Orchestrator has a narrower footprint. According to the Antigravity post, Orchestrator (Carbon) is rolling out for Google AI Ultra subscribers, while Pro subscribers get Orchestrator (Argon).
Pricing needs a careful read. Google's Gemini API pricing page lists Hydrogen at $1.50 and $7.50 per million tokens, with implicit context caching at $0.075 per million tokens and no explicit caching. Ars Technica describes the price as unchanged from 3.8 Flash. The pricing page shows the two match only from the new year.
| Model | Input per 1M tokens | Output per 1M tokens |
|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 |
| Gemini 3.8 Flash, through December 31, 2026 | $0.75 | $3.75 |
| Gemini 3.8 Flash, from January 1, 2027 | $1.50 | $7.50 |
For consumers, Google lists the Gemini app model drop-down. Ars Technica reports the model is live for AI Pro and Ultra subscribers there, while AI Plus and free users remain on an older model.
Pros and Cons
Pros
- The model combines a 1 million-token context window with up to 1 million output tokens at one flat price across the window.
- A four-step thinking dial, from low to xhigh, lets developers match reasoning depth to each task.
- Google reports large gains over 3.8 Flash, including 2.5x faster at the default setting and more than 5x fewer fabricated claims.
- Broad availability, from AI Studio and Android Studio to Search, means few access barriers for developers.
Cons
- Many of the speed, honesty and safety figures are Google-reported internal or partner-testing results rather than independent measurements.
- Access in the Gemini app is narrower than the headline suggests, because Ars Technica reports AI Plus and free users remain on an older model.
- The list price is twice the 3.8 Flash rate until December 31, 2026.
- The Carbon-paired Orchestrator option reaches only Ultra subscribers, and Google's two posts cite different Terminal-Bench versions for the result.
Outlook
Hydrogen reflects a design pattern in which a fast model does most of the work and a slower frontier model steps in for planning and recovery. Antigravity's Orchestrator makes that switching automatic, and Google contrasts it with static model routers that pick a model from the first prompt. Whether that dynamic approach holds up outside Google's own benchmarks is the open question. Independent testing of the safety and honesty figures, and of the partner-reported token savings, would show how much of the launch material carries over to real workloads. The scheduled change in 3.8 Flash's listed price on January 1, 2027 is also worth watching, since it alters the cost comparison.
Conclusion
Gemini 4 Hydrogen is a broad, fast general-purpose model with a flexible reasoning dial and a flat price over a very long context. Google's figures suggest large gains over 3.8 Flash, though most come from Google or its partners. It suits developers and teams who want an everyday coding and agent model, and Search users will meet it without choosing it. Teams with strict reliability needs should test it on their own workloads before relying on the published numbers.
Editor's Verdict
Gemini 4 Hydrogen: Fast Model, Four Thinking Levels earns a solid recommendation as an everyday execution model.
The strongest case for paying attention: the model combines a 1 million-token context window with up to 1 million output tokens at one flat price across the window. For long-document and agent workloads, the per-token price does not change as prompts grow. Reinforcing that, a four-step thinking dial, from low to xhigh, lets developers match reasoning depth to each task — practical value rather than just headline appeal. The broader signal worth registering is straightforward: the release pairs a fast executor with slower frontier models, so cost and latency are managed by routing work between models rather than by one model doing everything. On the other side of the ledger, one constraint is real rather than a marketing footnote: many of the speed, honesty and safety figures are Google-reported internal or partner-testing results rather than independent measurements. It should factor into any serious decision. Layered on top of that, access in the Gemini app is narrower than the headline suggests, because Ars Technica reports AI Plus and free users remain on an older model — a gap that matters more to consumers than to API users.
For developers on Gemini 3.8 Flash and teams building coding or agent workflows on the Gemini API, it is a serious evaluation candidate, with one caveat: 3.8 Flash stays at half the price until December 31, 2026, so test on your own workloads before switching.
Pros
- The model combines a 1 million-token context window with up to 1 million output tokens at one flat price across the window.
- A four-step thinking dial, from low to xhigh, lets developers match reasoning depth to each task.
- Google reports large gains over 3.8 Flash, including 2.5x faster at the default setting and more than 5x fewer fabricated claims.
- Broad availability, from AI Studio and Android Studio to Search, means few access barriers for developers.
Cons
- Many of the speed, honesty and safety figures are Google-reported internal or partner-testing results rather than independent measurements.
- Access in the Gemini app is narrower than the headline suggests, because Ars Technica reports AI Plus and free users remain on an older model.
- The list price is twice the 3.8 Flash rate until December 31, 2026.
- The Carbon-paired Orchestrator option reaches only Ultra subscribers, and Google's two posts cite different Terminal-Bench versions for the result.
References
Comments0
Key Features
1. Four thinking levels (low, medium, high as the default, and new xhigh); Google reports an Artificial Analysis Intelligence Index score of 51 at high versus 50 for Gemini 4 Argon, and 54 at xhigh. 2. Versus Gemini 3.8 Flash, Google reports 2.5x faster at the default setting, Terminal-Bench 4.1 of 57% versus 18.2%, and fabricated claims cut by more than 5x. 3. API list price of $1.50 per million input tokens and $7.50 per million output tokens, flat across the 1 million-token context window, with up to 1 million output tokens. 4. Orchestrator in Google Antigravity switches between Gemini 4 Carbon and Hydrogen in one session; Google reports near-Carbon task success with 24% less quota and 19% faster completion on average. 5. Google-reported agentic results include #5 on Agent Arena, 58.8% on HLE-Diamond, 52.3% on Vals Tax Agent Bench and 71.6% on Agents' Last Exam v1.1. 6. Rolling out across Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Google Search, Antigravity (paid AI Pro and Ultra), Android Studio and Gemini in Google Sheets.
Key Insights
- The release pairs a fast executor with slower frontier models, so cost and latency are managed by routing work between models rather than by one model doing everything.
- Google reports a higher Artificial Analysis Intelligence Index score for Hydrogen at default settings than for Argon, but that is a single index and not a full comparison of the two models.
- Most safety and honesty figures, such as overreach down by up to 81% and a 94% internal honesty rate, are Google-reported results from its own evaluations.
- Partner-reported speedups of 2x to 2.8x and 12% to 23% fewer tokens come from partners' early testing as reported by Google.
- Google's launch post cites Terminal-Bench 4.0 for the Orchestrator result while the Antigravity post cites Terminal-Bench 4.1, so the benchmark version should be checked before quoting it.
- Google's pricing page lists 3.8 Flash at half of Hydrogen's price through December 31, 2026, so the two match only from January 1, 2027.
- Search integration for all users globally gives Hydrogen far broader everyday exposure than the staged releases of the larger Gemini 4 models.
Was this review helpful?
Share
Related AI Reviews
Google AI Mode Adds Temporary Threads for Private Queries
Google's AI Mode in Search gets temporary threads: sessions detached from your account, not saved to history or used for personalization.
Gemini Tasks Goes Free: AI Calls, Payments, Hydrogen Model
Google makes Tasks in Gemini free in six countries on Gemini 4 Hydrogen; U.S. users add business calls, AP2 checkout and email or text intake.
Google Ask Maps Adds Trip Itineraries, Appointment Finding
Google's Ask Maps now builds day-by-day itineraries, adds route tips, finds appointments by calling businesses, and expands to 30 European countries.
Gemini Omni 1.2 Flash: 15-Second Video, 30% Lower Price
Longer 15-second clips, multi-turn editing and a $0.07-per-second 720p price headline Gemini Omni 1.2 Flash, now generally available.
