What Is Gemini 4 Hydrogen? Pricing, Specs and Benchmarks
Gemini 4 Hydrogen is Google's fast new model replacing Gemini 3.8 Flash: $1.50/$7.50 per 1M tokens, a 1M context window, four thinking levels, and it already powers Google Search and the Gemini app.
Quick Answer: Gemini 4 Hydrogen is Google's fast execution model, released November 12, 2026, as the successor to Gemini 3.8 Flash — $1.50/$7.50 per million input/output tokens, a 1M-token context window, four thinking levels, and it now powers Google Search and the Gemini app.
By the TechGamer App Review Team · Published November 13, 2026
What Is Gemini 4 Hydrogen?
Gemini 4 Hydrogen is the third model in Google's Gemini 4 family, launched in general availability on November 12, 2026. Google positions it as the speed tier: while the frontier models Gemini 4 Carbon and Gemini 4 Argon handle deep, long-horizon reasoning, Hydrogen is "built for rapid execution across diverse agentic workflows" — everyday coding, running tests, document work, and agent tasks.
It replaces Gemini 3.8 Flash, which launched in September, though Flash 3.8 stays available through the API for now (per Heise). Google dropped the "Flash" branding this time — 9to5Google notes the company chose the element name instead. Per the official model card, Hydrogen is adapted from Gemini 4 Carbon, and Google has not released the weights or parameter count.
| Spec | Value |
|---|---|
| Release (GA) | November 12, 2026, model ID gemini-4-hydrogen |
| Position | Fast execution tier of Gemini 4; succeeds Gemini 3.8 Flash |
| Context window | 1,048,576 tokens, unified input + output (Google Cloud docs) |
| Output | Text only; 64K default synchronous, up to 1M via background long decode |
| Inputs | Text, images, audio, video (about 45 min video with audio; up to 8.4 hours audio) |
| Knowledge cutoff | March 2026 (model card) |
| Pricing | $1.50 / $7.50 per 1M input / output tokens; $0.075 cached input |
| Thinking levels | Low, medium, high (default), xhigh |
How Much Does Hydrogen Cost?
Pricing is flat across the entire 1 million-token window: $1.50 per million input tokens and $7.50 per million output tokens, with cached inputs at $0.075 (per the official announcement). Ars Technica reports the price is unchanged from 3.8 Flash, though Heise calculates Hydrogen is effectively twice as expensive as Flash 3.8 at today's rates — Google is set to raise Flash pricing to the same level in January 2027.
The real cost lever is the thinking level. Independent tester Artificial Analysis measures an average cost of $0.29 per Intelligence Index task at low thinking versus $3.16 at xhigh — an 11x spread — with the xhigh tier costing 69% more than the default high for three extra index points. Budget-conscious users should default to high and reserve xhigh for genuinely hard problems.
Benchmarks: Google's Claims vs Independent Testing
Google reports large leaps over its predecessor. Treat the first table as vendor-reported numbers; the second is independent.
| Benchmark | Hydrogen | Comparison |
|---|---|---|
| AA Intelligence Index (high, default) | 51 | Gemini 4 Argon: 50 (Google-reported) |
| AA Intelligence Index (xhigh) | 54 | Argon: 50 (Google-reported) |
| Terminal-Bench 4.1 | 57% | Gemini 3.8 Flash: 18.2% (Google-reported) |
| Task speed vs 3.8 Flash | 2.5x faster | At high default (Google-reported) |
| Fabricated claims | Cut more than 5x | Vs 3.8 Flash (Google-reported) |
| HLE-Diamond | 58.8% | Leads its class (Google-reported) |
| Agents' Last Exam v1.1 | 71.6% | Tops other models (Google-reported) |
| Vals Tax Agent Bench | 52.3%, ranked #1 | Google-reported |
Artificial Analysis's independent run broadly confirms the strength but shows the trade-offs: at xhigh, Hydrogen scores 54, just behind Claude Sonnet 5.5 and GPT-6.1 Sol at 55 each. Its hallucination rate on the AA-Omniscience knowledge test is just 8%, against 43% for Sonnet 5.5 and 52% for GPT-6.1 Sol — but on command-line coding it trails, scoring 54% on Terminal-Bench 4.0 versus 64% for Sonnet 5.5. At the three lower thinking levels, AA found no tested model that is both more capable and faster.
| Thinking level | AA Index | Avg time per task | Avg cost per task |
|---|---|---|---|
| Low | 27 | 1.3 min | $0.29 |
| Medium | n/a | 3.7 min | between tiers |
| High (default) | 51 | 7.9 min | ~$1.87 |
| Xhigh | 54 | 13.7 min | $3.16 |
(Independent figures from Artificial Analysis; their published per-task cost points are $0.29 and $3.16 at the extremes, with time measured across Intelligence Index tasks.)
Independent Lab Results: Plotly and Vals AI
Two independent test labs published their own scores this week. Plotly's data-analytics exam gave Hydrogen a perfect 43 out of 43 — ahead of Gemini 4 Carbon (42) and Argon (41) — at roughly half the cost of either, with a median question time of 32 seconds against 93 for Carbon. Plotly's whole exam cost $6.36 on Hydrogen; it notes GPT-6.1 Sol matched the score for $2.61 (Plotly benchmark post).
Vals AI evaluated Hydrogen at xhigh thinking and ranks it #1 of 85 on Tax Agent Bench at 52.30%, ahead of Carbon (51.78%) and Claude Fable 5.1 (50.75%) — matching Google's own #1 claim. It also placed #2 of 40 on the Vals Finance Index (66.14%, behind Carbon), #6 of 37 on the Vals Index (68.49%, $9.84 per test), and mid-pack on coding: #11 of 64 on Terminal-Bench 4.0 at 54.04% (Vals AI model page).
| Independent test | Hydrogen result | Context |
|---|---|---|
| Plotly data-analytics exam | 43/43, median 32s per question | Carbon 42, Argon 41; Carbon took 93s; exam cost $6.36 |
| Vals Tax Agent Bench | #1 of 85 (52.30%) | Ahead of Carbon 51.78% and Claude Fable 5.1 50.75% |
| Vals Finance Index | #2 of 40 (66.14%) | Behind Carbon (67.05%) |
| Vals Index | #6 of 37 (68.49%) | $9.84 per test; leader Claude Fable 5.2 at 71.31% |
| Terminal-Bench 4.0 (Vals) | #11 of 64 (54.04%) | Mid-pack, consistent with AA's finding on CLI coding |
Known Limitations From the Model Card
Google's own model card lists a few caveats worth knowing before you migrate. Unjustified refusals are higher than 3.8 Flash, particularly on multimodal prompts, and the card acknowledges hallucinations, occasional slowness or timeout issues, and heavier token use at higher thinking levels. The knowledge cutoff is March 2026, and Google notes some domains may be less current. Ars Technica also points out the launch itself was unusually open: unlike the staged rollouts for Argon and Carbon, which began with cybersecurity researchers and initially locked general users to the most expensive subscription, Hydrogen went straight to broad availability (Ars Technica). Google's announcement adds that Hydrogen preserves 99.9% chain-of-thought legibility with zero deceptive reasoning edits across more than 200,000 audited turns (vendor-reported).
Where You Can Use It Today
Hydrogen rolled out globally on launch day across consumer, developer, and enterprise surfaces — the widest Gemini 4 rollout so far:
- Gemini app: available to everyone; free users get it through auto model selection when it fits the prompt.
- Google Search: global for all users — Google calls it "the biggest model upgrade to Search yet," and Search VP Robby Stein says it is the new default in AI Mode and handles the hardest AI Overviews questions (per The Verge).
- Developers: Gemini API and all tiers of Google AI Studio, plus Android Studio.
- Google AI Pro and Ultra subscribers: Google Antigravity, Chrome, and Gemini in Google Sheets.
- Via resellers: providers such as OpenRouter also list the model (per Plotly).
- Enterprises: Gemini for business via the API or model dropdown, and Sheets for eligible subscriptions.
A notable workflow feature ships alongside it: Orchestrator in Google Antigravity switches automatically between Carbon and Hydrogen inside one session. On Terminal-Bench 4.0 it reached near-Carbon success rates while consuming 24% less quota and finishing 19% faster on average (Google-reported). For parallel agent setups, Google suggests Carbon as coordinator and Hydrogen as worker and verifier agents.
Safety Changes Google Highlights
Because a fast model reaches more users doing real tasks, Google's announcement leans on agent-safety numbers: overreach in agentic tasks reduced by up to 81%, under-service by up to 79%, misaligned actions and task shortcutting cut 12x across tool-use and coding evaluations, up to 2x lower likelihood of bypassing permission boundaries on impossible tasks, and a 94% honesty rate on Google's internal agent honesty evaluation (all vendor-reported). On third-party testing, MLQ.ai's summary cites more than 2.3x resilience on the Gray Swan indirect prompt injection benchmark and false refusals under 0.4% on HealthBench Pro.
Should You Switch From Gemini 3.8 Flash?
For API users, the numbers argue for migrating: 2.5x faster tasks, Terminal-Bench 4.1 more than tripled, and far fewer fabricated claims, at a price that converges with Flash's anyway in January 2027. Google also quotes early partners seeing 2x to 2.8x faster runs on data-heavy enterprise workflows with 12% to 23% fewer tokens (vendor-reported). Keep Flash 3.8 only if you depend on features Hydrogen lacks — for example, the Google Cloud docs list RAG Engine and tuning as not supported for Hydrogen. For everyday users, there is nothing to do: Search and the Gemini app pick Hydrogen up automatically.
Frequently Asked Questions
What is Gemini 4 Hydrogen?
Google's fast execution model, released November 12, 2026 - the third Gemini 4 model, adapted from Gemini 4 Carbon and replacing Gemini 3.8 Flash for everyday coding and agentic work.
How much does Gemini 4 Hydrogen cost?
$1.50 per million input tokens and $7.50 per million output tokens, flat across the 1M-token window, with cached input at $0.075.
Is Gemini 4 Hydrogen free to use?
Yes, in the Gemini app for free users via auto model selection, and in Google Search for everyone. API and AI Studio access is pay-as-you-go.
Is Hydrogen better than Gemini 4 Carbon and Argon?
Not for deep reasoning - Carbon and Argon remain the frontier models. But Hydrogen scores above Argon on the AA Intelligence Index at its high and xhigh settings (51 and 54 vs 50) while finishing faster, per Google.
What replaced Gemini 3.8 Flash?
Hydrogen. Flash 3.8 remains on the API for now, with its prices set to rise to Hydrogen's level in January 2027.
What are Hydrogen's thinking levels?
Four: low, medium, high (default), and the new xhigh. Independent testing shows costs from $0.29 to $3.16 per task and times from 1.3 to 13.7 minutes across the levels.
Does Hydrogen have a 1 million token context window?
Yes - a unified 1,048,576-token limit shared between input and output, with up to 1M output tokens via background long decode.
Disclosure
This article is independent coverage based on Google's official announcements, the model card, Google Cloud documentation, and the independent and news sources listed below. Vendor-reported benchmark figures are labeled as such. It contains no affiliate links and no sponsorship.
Sources
- Google Blog – Introducing Gemini 4 Hydrogen for Agentic Workflows
- Google DeepMind – Gemini 4 Hydrogen Model Card
- Google Cloud – Gemini 4 Hydrogen documentation
- Artificial Analysis – Gemini 4 Hydrogen release analysis
- Heise – Gemini 4 Hydrogen: Google's efficient alternative to Argon and Carbon
- 9to5Google – Google announces Gemini 4 Hydrogen
- VentureBeat – Google launches Gemini 4 Hydrogen
- The Verge – Google Search AI Mode gets Gemini 4 Hydrogen
- Plotly – Gemini 4 Hydrogen Benchmark Results for Data Analysis
- Vals AI – Gemini 4 Hydrogen Benchmarks, Cost and Capabilities
- Ars Technica – Google rolls out Gemini 4 Hydrogen, finally replacing 3.8 Flash
- MLQ.ai – Google releases Gemini 4 Hydrogen