AnalysisNov 20, 20266 min read6 sources

Gemini 4 Neon's model card says it is adapted from Carbon; 'builds on Hydrogen' is Vertex listing copy

In one minute

Every technical document Google has published for Neon points to Carbon as the parent model. The Hydrogen wording appears in a Vertex AI Model Garden blurb, and Google has not reconciled the two.

Google released Gemini 4 Neon on 18 November 2026. It is the fourth Gemini 4 model, placed between the flagship Carbon and the fast Flash-tier Hydrogen. The documents Google published that day disagree on where the model comes from. Google DeepMind's model card says Neon was adapted from Carbon. The Vertex AI Model Garden listing, as quoted in independent coverage, says it builds on Hydrogen with deeper reasoning.

Lineage matters to builders for practical reasons. It tells you which model card's architecture, training-data and safety notes carry over. It hints at how Neon will behave next to the model it replaces. It also matters for governance records that require you to document a model's parent. We checked every Google primary page we could open to see which description the technical documentation supports.

What we checked

  • What does Google Cloud's model page say about Gemini 4 Neon's lineage, and has Google reconciled that description with the DeepMind model card? partly confirmed. The Vertex AI Model Garden listing, as quoted by OfficeChai, says Neon 'builds on Gemini 4 Hydrogen with deeper reasoning'. We could not open that console page directly. The public Vertex AI docs page for gemini-4-neon (last updated 2026-11-18) makes no lineage claim. We found no Google statement reconciling either text with the model card, which says Neon is adapted from Carbon. [1][2][4]
  • Does any other Google primary source (launch blog, AI Studio docs, Antigravity blog) back the Hydrogen lineage? refuted. No. The launch blog, written by Koray Kavukcuoglu, treats Carbon as Neon's baseline. The Antigravity post and the AI Studio docs mention Hydrogen only as a sibling tier, an executor partner or a comparison point. [3][5][6]

What we found

  • The Hydrogen wording appears in the Vertex AI Model Garden listing. OfficeChai attributes the sentence to Google's Vertex AI model page for Neon. The listing sits behind a Google Cloud console login and would not load for us, so we have not seen the sentence on a Google page ourselves. [1]
  • The public Vertex AI documentation page for gemini-4-neon is marked GA, lists a release date of 18 November 2026 and was last updated on 2026-11-18. It makes no lineage claim. It lists how Neon differs from earlier Gemini models but never says Neon builds on Hydrogen or was adapted from Carbon. The conflict therefore sits in the Model Garden marketing blurb, not in Google Cloud's reference docs. [2]
  • Google's launch blog, by Koray Kavukcuoglu, does not state what Neon is built from. Where it uses the word 'builds', it measures Neon against Carbon: in its agentic-behavior evaluation, Neon builds on Carbon's baseline. Hydrogen appears only as the fast-everyday tier. [3]
  • Hydrogen shows up in Google's other Neon pages as a comparison point or a partner model, never as a parent. The model card's safety section benchmarks Neon against Hydrogen. The Antigravity post pairs Neon as planner with Hydrogen as executor. This may explain the Model Garden wording: 'builds on Hydrogen' reads like tier positioning (Hydrogen's speed plus more reasoning), not a statement about which weights Neon came from. [4][5]

What it means

If you need a documented parent model, for an AI register, a vendor risk review or EU AI Act style records, cite the DeepMind model card: Neon is adapted from Gemini 4 Carbon. It is the only Google document that makes a lineage statement in a technical section. It defers architecture and training-data details to the Carbon card, so that card is the one to read. Treat the Vertex AI Model Garden line about building on Hydrogen as product positioning until Google says otherwise. Don't assume Neon inherits Hydrogen's training data or safety profile. The card benchmarks Neon's safety against Hydrogen; it does not derive it from Hydrogen.

On the practical side, the Hydrogen framing does match how Google wants Neon used: as a faster, cheaper step down from Carbon, and as a planner that hands work to Hydrogen executors in Antigravity. Run your own evaluations if you are switching from Carbon or Hydrogen. Google's descriptions of the lineage don't settle how Neon will behave on your tasks.

Competing readings

  • Neon is technically derived from Carbon. The Vertex 'builds on Hydrogen' line is loose positioning copy (Hydrogen's speed tier plus deeper reasoning), not a statement about weights (likely, about 72%). What would tell it apart: The model card's dependencies section names Carbon and defers architecture to the Carbon card [4]; The launch blog uses Carbon as Neon's baseline, and Hydrogen appears only as a sibling tier, a comparison point or an executor [2][3][4].
  • Neon really does derive from Hydrogen, scaled up with more reasoning, and the model card's Carbon dependency is the error (very unlikely, about 12%). What would tell it apart: The Vertex AI listing, as quoted by OfficeChai, says Neon builds on Hydrogen with deeper reasoning [1]; Neon shares Hydrogen's four thinking levels and its role as a fast tier, but no technical document states a Hydrogen dependency [2].
  • Neon draws on both models, for example distilled from Carbon onto a Hydrogen-sized base, and each document describes a different half (very unlikely, about 16%). What would tell it apart: The card says only 'adapted from' Carbon, without naming a method, which leaves room for a mixed lineage [4][1].

Base rate

The New Runtime archive could not be queried for comparable cases (canceling statement due to statement timeout), so it gives 0 cases: no base rate. The New Runtime archive returned 0 cases and 0 matches for comparable stories, which is too few to say anything. This conclusion rests on Google's own documents alone.

What would change this view

  • Google edits the DeepMind model card to name Hydrogen as Neon's dependency, or publishes a technical report describing Neon as derived from Hydrogen (model-listing signal): The Carbon dependency was the error, and Hydrogen's model card would become the reference for Neon's architecture and training data.
  • Google changes the Vertex AI Model Garden listing to drop or rephrase 'builds on Gemini 4 Hydrogen' (signal): This would confirm that the Hydrogen wording was positioning copy and that the model card is authoritative.

Still open

  • Will Google correct or align the Vertex AI Model Garden wording with the model card? We could not open the console listing directly to check its current text.
  • Is 'adapted from Carbon' a distillation, a fine-tune or a smaller variant? The card gives no method.
  • Has Google published a technical report for Neon that describes its training, beyond pointing to the Carbon card?

Evidence / sources

  1. 1officechai.com/ai/gemini-4-neon-to-be-googles-fast-reasoning-model-close-to-gemini-4-carbon-at-a-lower-price-and-faster-speed
  2. 2docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/4-neon
  3. 3blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon
  4. 4deepmind.google/models/model-cards/gemini-4-neon
  5. 5antigravity.google/blog/gemini-4-neon-in-google-antigravity
  6. 6aistudio.google.com/docs/models/gemini-4-neon
All Analysis

Next reads

Read next

Open the graph
  1. 01topicGoogleExplore the google topic hub.
  2. 02topicGeminiExplore the gemini topic hub.
  3. 03archiveAnalysisOpen the latest authorial analysis.
  4. 04sourcesSourcesSee the original sources this archive cites.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract