---
title: "Gemini 4 Sodium \"Drops Tomorrow\"? A Satirical Thread Meets Google’s Actual Lineup"
description: "Prime Intellect researcher @willcb joked that Gemini 4 Sodium drops tomorrow—and it got translated as real breaking news. Looking at Google's four actual Gemini 4 releases in seven weeks, it is easy to see why the parody landed."
url: https://codingpass.ai/blog/gemini-4-sodium-rumors-and-lineup
language: en
updated: 2026-11-20
---
# Gemini 4 Sodium drops tomorrow: when the real lineup reads like satire

@willcb joked on X that DeepMind is dropping Gemini 4 Sodium at $11.75/M as its best model for 'various things and stuff,' and the thread got translated literally as a leak. Google's real seven-week sprint of four element codenames and inverted pricing already reads like a parody.

[CodingPass team](https://codingpass.ai/about) Published Nov 20, 2026 Updated Nov 20, 2026 4 min read

If you opened your API console tomorrow and saw a brand-new `Gemini 4 Sodium`, you could almost be forgiven for shrugging and updating your config.

On November 19, Prime Intellect researcher Will Brown (`@willcb`) posted a deadpan joke on X claiming DeepMind would drop "Gemini 4 Sodium" the next day for `$11.75/M` (rising to `$16.50` soon) as "their best model for various things and stuff." Within a day, the thread had been translated literally onto Chinese social feeds as breaking news.

**For the record:** `Gemini 4 Sodium` and the "unreleased Helium" mentioned in the thread do not exist—`@willcb` made both up. The only reason the parody fooled anyone is that Google's real Gemini 4 lineup over the past seven weeks already reads like an Onion headline.

## Why the joke sounded plausible

`@willcb` followed up his opening post with two more "details": mid-tier Neon remains "their best and most expensive at `$15`," Hydrogen is `$7.50`, and flagship Carbon is `$10`; meanwhile Neon is based on Hydrogen and Carbon, Carbon is based on Argon, Argon is based on an unreleased Helium, and Sodium was trained concurrently from Neon and Helium.

In English, "various things and stuff" gives the joke away immediately. Strip the dry sarcasm out in translation, though, and `$15`, `$7.50`, and `$10` all match real numbers on Google's pricing page:

- **Neon at `$15` output**: Shipped November 18 as the "mid-tier" option between Carbon and Hydrogen, priced at `$3` input / `$15` output per million tokens
- **Hydrogen at `$7.50` output**: Shipped November 12 as the fast, lightweight tier, priced at `$1.50` input / `$7.50` output
- **Carbon at `$10` output**: Shipped November 10 as the flagship, normally `$4 / $20`, but discounted to `$2 / $10` through November 30

In other words, the "mid-tier costs more than the flagship" punchline was not invented—Google shipped that part itself.

## How messy the real lineup actually is

Line up what Google actually did over the last seven weeks and it is hard to blame anyone for losing track of the periodic table:

- **Argon disappeared from the app**: `Argon` launched in October for code and analysis, only to get yanked from the Gemini app model picker (alongside 3.1 Pro) the day `Neon` launched on November 18, while quietly remaining on the API pricing page
- **The mid-tier beats the flagship**: `Carbon` is billed as the heavy-reasoning flagship at `60` on the Artificial Analysis Intelligence Index, yet mid-tier `Neon` arrived eight days later scoring `61` at less than half the per-task cost and time
- **Inverted promo pricing**: Until November 30, flagship `Carbon` (`$2 / $10`) is cheaper per token than mid-tier `Neon` (`$3 / $15`), before jumping to `$4 / $20` on December 1
- **Rapid-fire element drops**: Google's release notes state `Neon` was adapted from `Carbon` (not Hydrogen plus Carbon, and with no Helium in sight), while averaging a new element every two weeks

| Element | Actual status | Official API rate (input / output per 1M tokens) | AA Intelligence Index | Where things stand |
| --- | --- | --- | --- | --- |
| Argon | Shipped (October) | Same as Carbon | — | Still live on the API; quietly dropped from the Gemini app picker |
| Carbon | Shipped (November 10) | $4 / $20 ($2 / $10 through Nov 30) | 60 | Supposed to be the flagship, currently cheaper and 1 point behind mid-tier Neon |
| Hydrogen | Shipped (November 12) | $1.50 / $7.50 | 54 | The one tier with an obvious job: fast, cheap edits and sub-agents |
| Neon | Shipped (November 18) | $3 / $15 ($0.15 cached input) | 61 | Pitched between Carbon and Hydrogen, somehow leads both in score and output price |
| Helium / Sodium | Parody only | $11.75 (fictional) | — | Invented by @willcb; leaves 114 periodic-table elements for next week's rumors |

## What to put in your config instead

Satire aside, your agent config still needs real model IDs. Regardless of which element Google pulls off the periodic table next, the four shipped models split cleanly enough in practice:

- **Everyday edits and sub-agents**: `Hydrogen` (`$1.50 / $7.50`) gives the lowest per-token cost and fastest turnaround
- **Multi-file refactors and long-running agents**: `Neon` at its default `high` thinking level scores `61` on the AA Intelligence Index at under half the per-task cost and time of Carbon
- **Heavy reasoning before November 30**: `Carbon` (`$2 / $10` intro rate) is temporarily cheaper per token than Neon until its `$4 / $20` rate kicks in

All four shipped models (`gemini-4-argon`, `gemini-4-carbon`, `gemini-4-hydrogen`, and `gemini-4-neon`) are live on CodingPass Go, Plus, and Pro, metered against your monthly allowance at official API rates.

Written by CodingPass team

We run CodingPass and test each guide against the latest stable release of the tool before we publish it. Spotted something out of date? Write to [support@codingpass.ai](mailto:support@codingpass.ai).

## Related reading.

Models

### [All models](https://codingpass.ai/models)

Every model's ID, minimum plan and monthly allowance on Go, Plus and Pro.

Benchmarks

### [Gemini 4 Neon benchmarks and pricing](https://codingpass.ai/blog/gemini-4-neon-benchmarks-pricing-thinking-levels)

Official and third-party benchmarks, cost per task across thinking levels, and monthly bill estimates.

Reference

### [How allowances and caps work](https://codingpass.ai/usage-limits)

The full rulebook: every allowance and cap, per plan and model, and the 429 response.
