🔍 Read the full analysis: Why Gemini 4 Neon’s Lower Bill Comes With Questions About The Data on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Google released Gemini 4 Neon on November 18, describing it as a balance of depth and efficiency. Artificial Analysis measured a one-point Intelligence Index gain over Gemini 4 Carbon at the models’ highest listed settings, alongside lower cost and shorter task times. Google’s model card says Neon is adapted from Carbon, but sends readers to Carbon’s documentation for architecture and training-data details; the full basis for those details remains unclear.
Google released Gemini 4 Neon on November 18, presenting it as a faster, more efficient model in its Gemini 4 lineup. Artificial Analysis measured Neon’s highest-effort setting one point above Carbon’s high setting on its Intelligence Index, while estimating lower cost per task and shorter completion times; Google’s model card says Neon is adapted from Carbon and points to Carbon’s documentation for details about its architecture and training data.
On Artificial Analysis’ Intelligence Index, Neon at xhigh effort scored 61, compared with 60 for Carbon at high effort. At those settings, the evaluator estimated an index task would cost $6.44 with Neon, against $7.76 with Carbon at its standard list price, and take 21.5 minutes rather than 37.1. These results compare different effort settings, so they do not isolate the effect of the model alone.
At the same high setting, Neon scored 60 and was estimated to cost $3.74 per index task. Artificial Analysis reported a cost of $1.90 at medium and $0.75 at low, with index scores of 55 and 44 respectively. Moving from high to xhigh added one point in this test and raised the estimated task cost from $3.74 to $6.44. Artificial Analysis described the release as “a small Intelligence gain over Gemini 4 Carbon at a ~17% lower Cost per Task.” That cost comparison uses standard list prices.
Google lists Neon at $3 per million input tokens and $15 per million output tokens. Carbon’s standard rates are $4 and $20, though Google’s stated promotion discounts Carbon to $2 and $10 through November 30. Artificial Analysis also reports that Neon produces output at about 110 tokens per second, roughly twice Carbon’s reported rate, but uses about 15% more output tokens per task. Its cached-input rate is $0.15 per million tokens, which Artificial Analysis describes as a 95% discount.
Gemini 4 Neon
Four models in seven weeks
Google’s Gemini 4 releases side by side, each at its best tested setting in the independent Artificial Analysis index. The best value in each row is outlined.
- Four models in seven weeksArgon on September 30, Carbon on October 28, Hydrogen on November 12, Neon on November 18. Neon is adapted from Carbon.
- A small step in the indexNeon at xhigh scores 61, one point above Carbon at high. It ranks third among model families, behind GPT-6.2 Astra and Claude Fable 5.2.
- A bigger step in cost and timeOne index task costs $6.44 instead of $7.76 and takes 21.5 instead of 37.1 minutes. Neon’s tokens are 25% cheaper than Carbon’s standard price.
- What to watchUntil November 30, Carbon still sells at a discounted $2/$10. Google’s own table compares Neon with Claude Opus 5.5 and GPT-6.2 Sol, not with the index leaders.
Source: Artificial Analysis, Gemini 4 Neon model page and article, accessed Nov 19, 2026 (Intelligence Index v5.0.1, cost and time per task at standard prices, release dates); Google, “Introducing Gemini 4 Neon”, Nov 18, 2026; Gemini 4 Neon model card. Output price in US$ per million tokens, standard rates. Redrawn by Thorsten Meyer AI.
Graphic: Powered by Thorsten Meyer AI
Lower Costs Depend on Effort Settings
The release matters to teams choosing a model for recurring work because a small benchmark difference can sit alongside a larger change in estimated cost and completion time. But the figures do not establish that Neon will be cheaper for every use. The estimates depend on task mix, output volume, effort setting and pricing; Carbon’s temporary discount also changes the comparison through November 30.
Neon’s coding results offer another reason to test the model against a specific workload rather than rely on a single overall ranking. Artificial Analysis put Neon at 58.6 on its Coding Agent Index, close to Claude Fable 5.2 at 58.5 and Claude Opus 5.5 at 58.2. Its estimated run cost was $6.86, compared with $15.47 for Fable 5.2 and $12.61 for Opus 5.5. Those are evaluator-specific results, not a promise of savings in a buyer’s own environment.
Performance also varied across individual tests. Neon at xhigh scored 63.6% on Terminal-Bench 4.0, below Carbon at high with 65.2%. On Artificial Analysis’ AA-Omniscience test, Neon’s reported hallucination rate was 14.4%, compared with Carbon’s 11.4%. The different results underline why benchmark scores, reliability measures and price need to be assessed separately before a model is deployed for consequential tasks.
Popular right now

Dyson Airwrap i.d. Multi-Styler with Bluetooth Connectivity

Dyson Cyclone V10 Origin Cordless Vacuum

TCL F8D Series 75-Inch 4K QLED Smart Fire TV, 144Hz, HDR (75F8D, 2026)
As an affiliate, we earn on qualifying purchases.
A Fourth Gemini 4 Model in Seven Weeks
Neon is the fourth Gemini 4 model Google has released in about seven weeks: Argon on September 30, Carbon on October 28, Hydrogen on November 12, then Neon on November 18. Google describes the lineup this way: Carbon is for difficult, long-running projects, Hydrogen for fast everyday execution, and Neon for a balance of depth and efficiency.
The model card adds that Neon is adapted from Gemini 4 Carbon. It does not set out separate architecture, training-data or hardware details in the material described by the source; instead, it directs readers to Carbon’s documentation. Neon accepts text, images, audio and video, returns text, and has a one-million-token context window. Google lists a March 2026 knowledge cutoff, four effort levels from low to xhigh, and high as the default.
Google’s evaluation document reports Neon as the best model in 16 of 19 listed rows, but compares it with Claude Opus 5.5 and GPT-6.2 Sol. It does not include Claude Fable 5.2 or GPT-6.2 Astra, which rank ahead of Neon on the Artificial Analysis Intelligence Index. In Google’s table, Claude Opus 5.5 leads Neon on Terminal-bench 4.1 and FrontierCode 2.0, while GPT-6.2 Sol leads on Chartography. The differing test sets mean the vendor table and independent index answer different questions.
Training Data Details Remain Limited
The model card’s link to Carbon documentation leaves readers without a distinct account, in the cited material, of what training data Neon uses or how its adaptation from Carbon changes the model. The available information also does not establish whether Neon has separate training or evaluation characteristics beyond those stated in Google’s documentation. No claim about undisclosed or improperly used data follows from this gap; it is a limit on what the cited material confirms.
Benchmark comparisons also have limits. Google’s table uses a different comparison set from Artificial Analysis’ index, and the supplied material does not provide enough detail to reconcile every difference in evaluation setup. Google says Neon had its “lowest fabrication rate yet in internal Google Antigravity testing,” but that is an internal result that outside readers cannot independently check from the material cited here. Real-world costs and reliability will depend on use, settings and actual usage volumes.
Access and Testing Will Shape the Picture
Neon is available through the Gemini API in Google AI Studio and Gemini for business. Google also lists access in Google Antigravity for paid AI Pro and Ultra subscribers, and in the Gemini app for Pro and Ultra subscribers. Google said a cyber-defense version, Gemini 4 Neon for cyber, would reach eligible members of its Fairwind Program “over the next few days.”
The next useful evidence will come from direct testing across matched tasks, model settings and pricing conditions. Teams comparing Neon with Carbon should account for Carbon’s discount through November 30, then compare costs using their own input and output volumes. Further documentation or independent evaluations could clarify how Neon’s adaptation relates to Carbon and whether its reported cost and speed advantages hold across broader workloads. The source material does not specify a date for any additional data or architecture disclosures.
Key Questions
What changed with Gemini 4 Neon?
Google released Neon on November 18, 2026, positioning it between Carbon and Hydrogen. Artificial Analysis measured a one-point Intelligence Index gain over Carbon at the compared high settings and reported lower estimated task costs and shorter completion times at the settings it tested.
Is Neon cheaper than Carbon?
Google’s standard listed rates for Neon are $3 per million input tokens and $15 per million output tokens, below Carbon’s standard $4 and $20. Artificial Analysis also estimated lower cost per index task for Neon in its tests. Carbon is discounted to $2 and $10 through November 30, so the price comparison depends on when and how the model is used.
What is known about Neon’s training data?
Google’s model card says Neon is adapted from Gemini 4 Carbon and refers readers to Carbon’s documentation for architecture and training-data information. The cited material does not provide a separate detailed account of Neon’s training data.
Does Neon outperform the leading models?
Not across every comparison. Neon ranked third among model families on Artificial Analysis’ Intelligence Index, behind GPT-6.2 Astra and Claude Fable 5.2. Google’s own table compares Neon with Claude Opus 5.5 and GPT-6.2 Sol instead, and shows rivals ahead in three of its 19 rows.
Where can people access Gemini 4 Neon?
Google lists Neon in the Gemini API through Google AI Studio and in Gemini for business. It is also available in Google Antigravity and the Gemini app for paid AI Pro and Ultra subscribers, according to the supplied launch information.
Source: ThorstenMeyerAI.com
