Google has not announced Gemini 4 Flash. As of November 8, 2026, two codenames point at it: Hydrogen, which leak-watchers on X say Google has been testing on LMArena and in AI Studio since late October, and Hyperion, a new model family that turned up in builds of Cognition’s Devin, listed just before Gemini 4 Argon.
Neither name comes from Google or Cognition. Below: what each sighting shows, what Google has actually said, whether Hydrogen and Hyperion are the same model, and what would settle it.
On this page · 9 sectionsOpen
Google has not announced Gemini 4 Flash; its November 2 recap names Gemini 4 Argon and Carbon only.
Hyperion, a new model family with minimal to high effort levels, appeared in Devin builds listed just before Gemini 4 Argon, per two trackers on November 8.
Hydrogen, reported as Gemini 4 Flash’s codename since October 27, is said to be in LMArena and AI Studio tests at 300 to 500 tokens per second; no one has measured that independently.
The November 18 release estimates on X come with no source; Google has given no date.
Google’s current Flash model, Gemini 3.8 Flash, costs $0.75 and $3.75 per million tokens through December 31, 2026.
§ 01What we actually know
| Claim | Source | Status |
|---|---|---|
| Google is shipping more Gemini 4 models | Google’s October recap, November 2 | Confirmed for Argon and Carbon; no Flash model named |
| Gemini 4 Flash is codenamed Hydrogen | X posts since October 27, collected by OfficeChai | Reported, not confirmed |
| Five Hydrogen checkpoints at 300 to 500 tokens per second | X posts, November 4 | Reported, not measured independently |
| Hydrogen is in A/B tests on LMArena and AI Studio | X posts, November 4 | Reported |
| A Hyperion model family sits in Devin builds | X posts, November 8 | Seen by two accounts; nothing from Cognition |
| Hyperion is Gemini 4 Flash | Its place in the model list | A guess |
| Release around November 18 | Two X accounts | No source given |
§ 02What Google has said
Google’s October roundup, published November 2, covers Gemini 4 Argon and Gemini 4 Carbon. It says the company “released one frontier model globally and announced our next one just two weeks later.” It names no Flash model and no Hydrogen.
Google’s newest Flash model is still Gemini 3.8 Flash, released September 2 at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, then $1.50 and $7.50. Our Gemini 3.8 Flash record has the detail, and our Gemini 4 page tracks Argon and Carbon.
§ 03Hydrogen: the arena sightings
OfficeChai traced the name to an October 27 post by Pankaj Kumar: “Gemini 4 Flash is in internal testing under the codename Hydrogen.” On November 4, two more accounts said five new Hydrogen checkpoints had appeared, with speeds of 300 to 500 tokens per second, and that the model was in A/B tests on AI Studio and LMArena. Google has not commented, and no one has published an independent speed measurement.
The most shared test came on November 6. One tester caught Hydrogen in an arena battle on a motion graphics prompt and timed it against Gemini 3.8 Flash on the same prompt.
That is one prompt and one run each, timed by hand, so read it as a sign of speed rather than a benchmark.
§ 04Hyperion: the Devin sighting
The second name surfaced on November 8. An account that tracks Devin builds went through Saturday’s desktop build, 3.12.1028, and wrote that hyperion was “new since the Oct 30 build” with minimal to high effort levels: “It’s listed among the Gemini models, right before Gemini 4 Argon, so possibly Google.” Its own verdict: nothing confirmed.
Five hours later another tracker reported the same family in Devin CLI v3000.12.17, with five variants, hyperion, hyperion-minimal, hyperion-low, hyperion-medium and hyperion-high, placed between Gemini 3.5 Flash Lite and Gemini 4 Argon: “This makes me think Hyperion could be Gemini 4 Flash although it’s still unconfirmed.” That is the post that spread.
What we could check ourselves: Cognition’s public model page lists Gemini 4 Argon and Gemini 3.7 and 3.8 Flash, and no Hyperion. Its SDK guide names a bundled CLI build, 3000.11.1, and we found no release note for 3000.12.17. The same build also carries a second unattributed codename, wyvern, with low, xhigh and max effort levels.
§ 05Are Hydrogen and Hyperion the same model?
Possibly, but nothing ties them together yet. The position fits: Hyperion sits among Devin’s Gemini entries, below Argon, where a Flash model would go. The name does not: Google has named its Gemini 4 models after elements, Argon and Carbon, and Hydrogen fits that pattern; Hyperion does not. Hyperion may be a label Cognition gave to a preview it is testing, as other codenames in Devin builds this week appear to be. If it is a Google model, it could also be something other than Flash.
§ 06When could it ship?
There is no date. The November 18 estimates on X come with no source. For pacing, Google released Gemini 4 Argon on September 30 and announced Gemini 4 Carbon on October 28. A model showing up in arena tests and in a partner’s build is often a sign that a launch is weeks away, not months, but Google has not said so.
§ 07Where CellCog sits
Our conflict, declared: every CellCog tier runs on Claude Opus 5.5, Anthropic’s model, and we do not route to Gemini today. We track Flash-class launches because a fast, cheap model resets what every agent product can afford to do in a loop, ours included.
§ 08What would settle it
- Google naming it. A Google or DeepMind post, or a gemini-4-flash model id in the Gemini API docs or AI Studio.
- A listing elsewhere. An OpenRouter or Vertex AI entry with matching speed and price.
- Cognition. A Devin release note naming Hyperion’s maker, or the model appearing on its public model page.
- LMArena. A published score for Hydrogen under its real name.
§ 09Sources
- Google: The latest AI news we announced in October, November 2, 2026.
- OfficeChai: Google Preparing To Release A Flash Model Named Gemini 4 Hydrogen, X Users Say, November 5, 2026.
- @aitrackerbot on X, November 8, 2026, 05:04 UTC.
- @Mr_Salio on X, November 8, 2026, 10:08 UTC.
- @deweytn on X, November 6, 2026, 20:54 UTC.
- Cognition: AI Models and Local Devin sessions with the SDK, read November 8, 2026.
Q1When will Gemini 4 Flash be released?
Google has given no date. Two X accounts estimate around November 18 without citing a source.
Q2Is Hyperion the same as Hydrogen?
Unknown. Hyperion’s place in Devin’s list fits a Flash model, but Google names its Gemini 4 models after elements and Hyperion breaks that pattern.
Q3How fast is Hydrogen?
X posts report 300 to 500 tokens per second. One tester timed a motion graphics prompt at about 2 minutes against about 11 for Gemini 3.8 Flash. Neither is an independent benchmark.
Q4What will Gemini 4 Flash cost?
Google has published no price. Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, then $1.50 and $7.50.
Q5Does CellCog use Gemini Flash?
Not today. Every CellCog tier runs on Claude Opus 5.5.