Gemini 4 Neon arrives as Google's cheaper middle tier, with a price edge on a timer

5 Sources

Share

Google launched Gemini 4 Neon on November 18, a new mid-tier model that, by Google's positioning, slots between the fast Hydrogen and the heavyweight Carbon in the Gemini 4 family, aimed at complex coding, analytical reasoning and enterprise work

4

. It is available the same day, with no preview period

4

.

Source: Interesting Engineering

Source: Interesting Engineering

Google says the model is meant to balance depth and efficiency, delivering frontier-level reasoning for complex, high-frequency work faster and more cheaply than either sibling

4

. It arrives a week after Carbon and Hydrogen

3

. Android Authority notes the tiered, multi-model approach mirrors what rivals such as Anthropic and OpenAI do, and Google believes Neon is a competitive alternative to their offerings

1

. Neon takes the middle slot Argon held; Google had pitched Argon as the "ideal choice for the majority of software engineering and knowledge work"

3

.

Source: VentureBeat

Source: VentureBeat

Pricing and access

Input tokens cost $3 per million and output tokens $15 per million, with cached input at $0.15 and a 1 million-token context window

4

. Per 9to5Google, Neon costs twice as much as Hydrogen

3

. Carbon and Argon are currently at promotional pricing of $2 and $10 per million tokens until November 30, after which they rise to $4 and $20

3

. That makes Neon 25% cheaper than Carbon's regular price

3

.

The bigger access change is who can buy it. Neon works with Google AI Pro plans, where Carbon is restricted to AI Ultra accounts

1

. The launch also reshuffles the model picker: 3.1 Pro is removed, Neon replaces Argon, and paid users get a four-option menu of Auto, Hydrogen, Neon and Carbon

3

.

The benchmark case

Google frames the results as evidence that Neon delivers frontier intelligence with fast end-to-end task completion

4

. Per Google, Neon scores 61 on one third-party index, the Artificial Analysis Intelligence Index, at its xhigh thinking level, ranking #3 overall just behind larger and slower models

4

. On coding, Google reports a #1 result on one coding agent index, the Artificial Analysis Coding Agent Index

4

. Its 39.0% SWE-sweep score holds only inside Google's own Antigravity harness, against 12.8% on the standard mini harness

4

.

The core pitch is the comparison with Carbon. In early internal evaluations, Google says, Neon matched or exceeded Carbon on overall quality while finishing tasks three times faster, at roughly 60% of the cost

2

. At its default high setting, Google puts Neon's cost per task up to roughly 80% below top competing frontier models on the evaluations it names

4

. The model is also designed to act as an orchestrator directing parallel Hydrogen subagents in Antigravity, and Google reports its lowest fabrication rate yet in its own internal testing

4

.

A gated security variant

A specialized "Gemini 4 Neon for cyber" version, per Google, writes more secure code than Carbon on every AutoBaxBench difficulty tier and posts state-of-the-art security benchmark scores at roughly half the cost of the larger model

4

. Initially, only trusted defenders get access, through Google's Fairwind Program

2

. Per 9to5Google, the promotional rates for Carbon and Argon run until November 30; the 25% saving applies to the regular prices that follow

3

.

© 2026 TheOutpost.AI All rights reserved