· via dev.to (home feed)
Google swaps Gemini version numbers for element names like Hydrogen and Neon
Google's Gemini 4 family trades decimal version numbers for element names — Hydrogen, Neon, Argon, Carbon — framing the models as lateral trade-offs between cost, speed and depth rather than sequential upgrades.

Element names replace version numbers
Google has dropped decimal version numbers for its Gemini 4 family and is shipping models under chemical element names instead. According to a post on dev.to by Evgenii Zinner, model pickers in IDEs and API clients have filled in quick succession with Argon, Carbon, Hydrogen and, most recently, Neon — leaving developers to work out which name corresponds to which job, and at what cost.
How the four models differ
Based on the pricing and behaviour described in the post:
- Gemini 4 Hydrogen is the lightweight option, listed at $1.50/$7.50 per million tokens on the fast quota pool. It is built for quick file edits, unit tests and terminal commands, and offers an "xhigh" thinking setting for harder algorithmic problems.
- Gemini 4 Neon is the newest release and the new default for Pro users. Zinner describes it as a distilled spin-off of Carbon that offers reasoning quality close to the flagship at roughly twice the speed, at $3.00/$15.00 per million tokens on the Pro quota pool.
- Gemini 4 Argon was the original Gemini 4 preview and the testbed for native Orchestrator mode. It has already been retired in favour of Neon.
- Gemini 4 Carbon is the flagship on the Ultra/enterprise tier, intended for large codebase audits, formal reasoning tasks and involved architectural refactors.
Why the naming changed
The post gives two explanations, both of which are the author's interpretation rather than statements from Google.
The first is release velocity. Checkpoints that once waited months for product packaging now go straight from DeepMind to Google AI Studio, Vertex AI and Antigravity, and the internal elemental codenames travel with them. Renaming endpoints such as gemini-4-hydrogen and gemini-4-neon would, the argument runs, cost weeks that competitors will not hand back.
The second is conceptual. Two decades of software versioning taught developers that a higher number replaces a lower one, and that staying on an older release amounts to technical debt. That model fits poorly with agentic workflows, where a single task chains several models together: putting a frontier model on a mechanical multi-file search-and-replace wastes budget, while a fast model drafting a migration plan misses edge cases and produces broken code. Element names, Zinner argues, signal that these are lateral trade-offs between cost, speed and depth — peers with different jobs, not steps on an upgrade path. Neon can plan while Hydrogen executes, and neither replaces the other.
Suggested defaults
For everyday coding, the post recommends defaulting to Hydrogen and flipping its thinking toggle to xhigh for difficult maths or algorithmic problems. For architecture and planning, it recommends Neon, or an Orchestrator mode that pairs Neon's planning turns with Hydrogen's execution speed. The broader advice is to treat the elements as specialists with distinct responsibilities rather than releases to keep up with.
Why it matters
Model selection is becoming role-based. If the element names stick, "which model should I use?" stops meaning "which is newest?" and starts meaning "which trade-off fits this step of the loop?" — a shift that touches IDE defaults, agent orchestration and how teams budget their token spend.
The naming also carries practical risk. Argon, the first name to appear, has already been retired, so pinning code to a specific element may mean chasing deprecations. And the underlying trend — internal research codenames pushed to production endpoints to keep pace with rivals — suggests model churn will keep accelerating rather than settling down.
One caveat worth flagging: this is a single developer's account. The role mapping, pricing figures and explanations above come from that post rather than official Google documentation, and Google has not, in this source, confirmed its reasoning for the periodic-table branding.
- #gemini
- #ai-models
- #llm
- #developer-tools