cd /news/artificial-intelligence/googles-gemini-4-lineup-splits-power… · home › topics › artificial-intelligence › article
[ARTICLE · art-172841] src=cryptobriefing.com ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

Google’s Gemini 4 lineup splits power, speed and price across four models

Google released the Gemini 4 family as four models — Argon, Carbon, Hydrogen and Neon — all sharing a unified 1M-token context window, with Carbon positioned as the most powerful at low effort and Neon and Hydrogen scaling better as reasoning effort rises. Carbon and Argon are priced at $2 per million input tokens and $10 per million output tokens through November 30, 2026, then double to $4 and $20; Hydrogen costs $1.50 input and $7.50 output, and Neon $3 input and $15 output, with output priced at five times input across the lineup. Carbon reportedly reaches approximately 65% on Terminal-Bench 4.0, and Neon matches Carbon's coding results at reduced time and cost, with the models available through the Gemini API, the Gemini app, Google Antigravity on paid plans, and Search and AI Mode, while Carbon is restricted to the Ultra tier in specific consumer services.

key takeaway

Carbon and Argon cost $2 per million input and $10 per million output tokens only through November 30, 2026, then double to $4 and $20, while Neon at $3/$15 matches Carbon's coding results at reduced time and cost.

by read4 min views4 publishedNov 19, 2026
Google’s Gemini 4 lineup splits power, speed and price across four models
Image: Cryptobriefing (auto-discovered)

gemini logo bg

Carbon leads on hard tasks at low effort, while Neon and Hydrogen are built to scale as reasoning effort climbs

Google’s Gemini 4 family is now fully out, and it does not come as one model. It comes as a small periodic table.

The lineup has four variants: Argon, Carbon, Hydrogen and Neon. The core message is that Carbon costs the most and performs best at low effort, while Neon and Hydrogen scale better when you ask them to think harder.

Four models, one context window #

All four Gemini 4 models share a unified 1M-token context window. A token is roughly a chunk of a word. A million of them means each model can hold a very large pile of text, code or documents in working memory at once.

Where the models differ is in reasoning depth and execution speed. Google has positioned each one for a different slice of work, from casual developer tinkering to enterprise-grade tasks.

The rollout came in stages:

  • Argon arrived first, on October 13, 2026.
  • Carbon followed on November 10, 2026.
  • Hydrogen landed on November 12, 2026.
  • Neon closed out the set on November 18, 2026.

Carbon is pitched as the most powerful option. It is tuned for complex problems that require deep reasoning, the kind of work where a wrong answer costs more than a slow one.

Neon targets intricate tasks too, but with an emphasis on better speed and cost-efficiency. Hydrogen sits at the practical end of the shelf. It is designed to execute everyday jobs quickly at a lower price point.

The pricing math #

Gemini 4 pricing is quoted per million tokens, split between input (what you send the model) and output (what it sends back). Output is priced at five times input across the board.

  • Hydrogen: $1.50 input, $7.50 output
  • Neon: $3 input, $15 output
  • Carbon and Argon: $2 input, $10 output through November 30, then $4 input, $20 output afterwards

That November 30 cutoff creates a quirk worth noticing. Until the end of the month, Carbon is actually cheaper per token than Neon. After that, Carbon’s rates double and it becomes the priciest model in the lineup.

Hydrogen’s output rate is exactly half of Neon’s. For workloads that generate a lot of text, that gap adds up quickly.

AI, tech, and the markets they move—in one daily briefing.

Daily. Free. Join 34,000+ readers across crypto, finance, and policy.

Effort is now a dial #

Gemini 4 models support adjustable reasoning effort: low, medium, high and, in some cases, xhigh.

More thinking generally means more computation per request. That typically means more output tokens and more time before an answer appears. The dial lets developers decide how much they want to spend on each problem.

Carbon is optimized for medium effort, and it performs best of the group at low effort. The other models are optimized for high effort. That is where Neon and Hydrogen start to close the gap and scale better.

Benchmarks and access #

On difficult tasks, Carbon reportedly achieves approximately 65% on Terminal-Bench 4.0. That benchmark tests how well models handle hard, terminal-style work, the sort of command-line tasks developers and agents run constantly.

In coding evaluations, Neon matches Carbon’s results at reduced time and cost.

The models are available across several Google surfaces:

  • The Gemini API, for developers building their own products
  • The Gemini app
  • Google Antigravity, on paid plans
  • Search and AI Mode, where the models are integrated directly

In specific consumer services, Carbon is restricted to the Ultra tier.

What this means #

For developers, the practical question is no longer just “which model is best.” It is “best at what effort level, and at what price per million tokens.” A coding team that can tolerate high reasoning effort might find Neon delivers Carbon-level results for less time and money, based on the coding evaluations. A team that needs strong answers fast, with minimal thinking overhead, has a clearer case for Carbon. That is especially true before November 30, while its rates sit below Neon’s.

Hydrogen’s role is more about volume. At $1.50 input and $7.50 output, it is the obvious candidate for routine, high-frequency jobs where speed matters more than depth.

Once Carbon and Argon move to $4 input and $20 output, some workloads that defaulted to Carbon may get rerouted to Neon or Hydrogen. That is particularly likely for workloads where higher effort settings can make up the performance difference.

The broader takeaway for anyone building on these tools is that cost optimization now has two axes. You choose a model, and then you choose how hard it should think.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @google 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/googles-gemini-4-lin…] indexed:0 read:4min 2026-11-19 · —