Google has not announced Gemini 4 Hydrogen, but on November 12 testers reported Vertex AI responses from gemini-4-hydrogen plus a Google Cloud billing entry. Neither is an announcement; both are worth checking directly.
Hydrogen is a Flash-tier leak, not a shipping SKU
Hydrogen is the reported codename for the Gemini 4 Flash generation — Google's fast, cost-efficient tier, traced by leak trackers to X posts starting October 27 (see the ArgoLink timeline). Google's own Gemini API model list still names only Gemini 4 Argon, with no Hydrogen entry. Google's latest official Gemini 4 announcement covers Gemini 4 Carbon, not Hydrogen.
What changed on November 12
| Signal | What was posted | Confidence |
|---|---|---|
| Billing SKU | @LuminaBench: "Gemini 4 Hydrogen is in the Google Cloud billing catalogue," pricing read $0.75 in / $3.75 out / $0.0375 cache; same figures from @0xdominus | Medium: screenshots, two accounts, no Google confirmation |
| Vertex access | @thtbee_: "Hydrogen is returning responses on Vertex API"; @Delroy715: HTTP 200 on global endpoint, ~150 tok/s, native Thinking on | Medium: detailed traces with prompts, project-scoped |
| Launch tease | Logan Kilpatrick (@OfficialLoganK) posted "Gemini" on Nov 12 (post) | Low: suggestive timing, not confirmation |
Verify it yourself before believing screenshots
Do not hard-code gemini-4-hydrogen into production. If you have GCP access, check directly.
- Search the Cloud billing catalogue or pricing export for
hydrogen. A billable SKU is a strong pre-launch hint, though catalogue rows alone never prove a launch date. - Try a minimal Vertex call with your project's auth and region, e.g.
POST /v1/projects/PROJECT/locations/global/publishers/google/models/gemini-4-hydrogen:generateContentwith a smallgenerateContentbody. Treat HTTP 200 as "enabled for this project" and any 404/403 as inconclusive — it can mean no allowlist, wrong region, or wrong ID, not proof the model exists. - Check Google's own model list and changelog. Hydrogen counts as released the day a
gemini-4-flashorgemini-4-hydrogenID appears in Google's own reference — until then, treat it as test-only.
What early tests show
Early creative-output tests are anecdotal, not a deployment signal — but one probe keeps coming up: the "pelican riding a bicycle" test, where Hydrogen's output differs visibly from prior Flash output.
"The pelican riding a bicycle test is basically solved. Gemini 4 Hydrogen (Flash) made one with its feet actually on the pedals, a basket up front, a lighthouse in the back and a live speed readout." — @AnikaCodes (real-user test with video, Nov 6)
Treat the viral 300-500 tok/s band as an unmeasured multi-checkpoint spread. The only grounded figures: roughly 150 tok/s in one detailed Vertex run (@Delroy715), versus OrcaRouter's Artificial Analysis read of about 113 tok/s for shipping 3.8 Flash and 68 for Argon.
How much Hydrogen will cost
Google has published no Hydrogen price; confirmed rows below come from Google's pricing page, the last row from leaked screenshots.
| Model | Input / 1M | Output / 1M | Status |
|---|---|---|---|
| Gemini 3.8 Flash | $0.75 ($1.50 after Dec 31, 2026) | $3.75 ($7.50 after Dec 31, 2026) | Generally available |
| Gemini 4 Argon | $2.00 ($4.00 from Dec 1, 2026) | $10.00 ($20.00 from Dec 1, 2026) | Generally available |
| Gemini 4 Carbon | $2.00 ($4.00 from Dec 1, 2026) | $10.00 ($20.00 from Dec 1, 2026) | API access from Nov 10 |
| Hydrogen (leaked) | $0.75 | $3.75 (+ $0.0375 cache) | Unconfirmed |
For a 50M input + 10M output token workload: about $75 at 3.8 Flash or leaked Hydrogen rates, versus $200 on Argon/Carbon intro rates and $400 after December 1 — token cost only; real task cost depends on retries and reasoning length. Do not budget the leak; December 1 (Argon/Carbon doubling) and December 31 (3.8 Flash doubling) are the only committed dates.
Where Hydrogen sits
If the leaked price and speed hold, Hydrogen would fit the bottom rung by price and latency — high-volume chat, tool loops, agent inner loops — leaving planner steps and hard reasoning to Argon and Carbon.
| Dimension | Hydrogen (reported) | Argon | Carbon | 3.8 Flash |
|---|---|---|---|---|
| Public API ID | None | gemini-4-argon | gemini-4-carbon | gemini-3.8-flash |
| Positioning | Fast Flash-like | Frontier coding/agentic | Cybersecurity, long horizons | Current Flash workhorse |
| Context | Unknown | 1M tokens | 1M-class | 1,048,576 tokens |
| Access Nov 12 | Billing + Vertex 200 (some projects) | Generally available | API since Nov 10 | Generally available |
Confirmed-row positioning, context, and availability reflect Google's model list, pricing page, Carbon announcement, and Carbon model card. Watch for collisions: "Hydrogen-1.0" is an unrelated open-weight science model and "Hyperion" a separate label seen in Devin builds with no confirmed Google link — both flagged as lookalikes by leak trackers (CellCog's Hyperion breakdown).
What to do this week
- Experiment only while access is project-scoped. Consider production only after Google lists a model ID, documents limits, and publishes pricing.
- Keep production on 3.8 Flash (volume) and Argon/Carbon (hard tasks); hold the model ID in a variable so a Flash successor is a one-line swap.
- Budget December 1 first: Argon/Carbon doubling to $4/$20 is confirmed; Hydrogen savings are not.
Related reading: Gemini 4 Argon API pricing and availability, Gemini 4 Carbon vs Argon guide, Gemini 4 Carbon API pricing review.
FAQ
Can I call Hydrogen from the Gemini API, Vertex, or AI Studio today?
Only project-scoped Vertex responses have been observed, for some GCP projects. No public API ID has appeared in Google's model list. Test with your own project; do not ship on it.
Is Hydrogen the same thing as Hyperion or Hydrogen-1.0?
No confirmed link to either. Hyperion is a separate label in Devin builds; Hydrogen-1.0 is an unrelated open model. Details in the section above.