AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Gemini 4 Neon And The Cost-Performance Question Google’s Table Raises on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

  • Fast, free delivery for dorm and study essentials
  • Prime Video and Amazon Music included
  • Member-only deals
Try Prime for Young Adults Free trial for eligible 18–24 year olds
As an affiliate, we earn on qualifying purchases.

TL;DR

Google released Gemini 4 Neon on November 18, presenting it as a balance of capability and efficiency. Artificial Analysis measured a one-point Intelligence Index gain over Gemini 4 Carbon at the highest settings, alongside lower per-task cost and shorter task times; results vary with effort level. Google’s published comparison table does not include the two models ahead of Neon in the independent index.

Google released Gemini 4 Neon on November 18, positioning it between its Gemini 4 Carbon and Hydrogen models. In measurements from Artificial Analysis, Neon scored 61 on the Intelligence Index at xhigh, one point above Carbon at high, while an index task cost less and took less time; the results are independent measurements, not a guarantee for every workload.

Artificial Analysis put Neon’s xhigh run at $6.44 per index task and 21.5 minutes, compared with $7.76 and 37.1 minutes for Carbon at high and its standard list price. At the same high effort level, Neon scored 60 and cost $3.74 per task. These figures compare specified configurations and standard prices; they do not establish that Neon will be cheaper or faster on every application.

Neon’s listed prices are $3 per million input tokens and $15 per million output tokens, against Carbon’s standard $4 and $20. Carbon is discounted to $2 and $10 through November 30, according to the supplied report, changing the near-term price comparison. Artificial Analysis also reports that Neon generates about 110 output tokens per second, roughly twice Carbon’s reported rate, while using about 15% more output tokens per task.

The effort setting changes the cost-capability balance. Artificial Analysis reports Neon at 61 points for $6.44 at xhigh, 60 for $3.74 at high, 55 for $1.90 at medium, and 44 for $0.75 at low. Google says Neon is adapted from Carbon. Its stated inputs include text, images, audio and video, with text output and a one-million-token context window; Google lists availability through several of its consumer, developer and business products.

At a glance
reportWhen: Released November 18, 2026; independent…
The developmentGoogle launched Gemini 4 Neon, a model adapted from Carbon, with independent measurements showing a small top-setting score gain and lower cost and completion time per index task.

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.

ArgonCarbonHydro.Neon ReleasedSep 30Oct 28Nov 12Nov 18 Index50605461 Cost/task$4.97$7.76$3.16$6.44 Min./task29.737.115.321.5 Output $$10$20$7.50$15 Carbon $2/$10 discount ends Nov 30Neon: +1 point, $1.32 less per task
  1. 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.
  2. 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.
  3. 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.
  4. 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

The Cost of More Model Effort

The release puts the practical question less on whether Neon is categorically smarter than Carbon and more on how much capability a workload needs. In Artificial Analysis’s figures, moving from high to xhigh adds one Intelligence Index point but raises the cost per index task from $3.74 to $6.44. For teams running many model calls, that configuration choice could matter more to budgets than the one-point headline difference.

Those results are not a universal price quote: task costs depend on the benchmark, model settings, token use and pricing conditions. The comparison also shows why raw model rankings are only part of a purchasing decision. On the Coding Agent Index, Neon scored 58.6, close to Claude Fable 5.2 at 58.5 and Claude Opus 5.5 at 58.2; the reported task costs were $6.86, $15.47 and $12.61 respectively. Independent test scores do not by themselves predict performance on a company’s own codebase or workflow.

For readers choosing a model, the table makes a case for testing both quality and cost at the intended effort level. A lower benchmark-task cost can be useful, but it does not settle questions of reliability, latency under production conditions, or the value of the model’s output in a specific application.

Ninja Prestige Smart XL 10-in-1 Air Fryer and Countertop Oven with Pro Cook System, Stainless Steel, DT551

Ninja Prestige Smart XL 10-in-1 Air Fryer and Countertop Oven with Pro Cook System, Stainless Steel, DT551

Vitamix 5-Speed Immersion Blender, 625-Watt, Stainless Steel/Black

Vitamix 5-Speed Immersion Blender, 625-Watt, Stainless Steel/Black

As an affiliate, we earn on qualifying purchases.

A Fourth Gemini Model in Seven Weeks

Neon is the fourth Gemini 4 model Google has released in roughly seven weeks: Argon arrived September 30, Carbon on October 28, Hydrogen on November 12, and Neon on November 18. Google describes Carbon as suited to difficult, long-running projects and Hydrogen to fast everyday execution, with Neon intended to balance depth and efficiency. The model card’s description of Neon as adapted from Carbon is relevant to interpreting the launch: this is a new offering, but Google points readers to Carbon documentation for architecture, training data and hardware.

Google’s launch post calls Neon a model for “frontier-level intelligence with fast end-to-end task completion.” Its evaluation document lists 19 comparisons and marks Neon best in 16. The rivals in that table are Claude Opus 5.5 and GPT-6.2 Sol. It does not include Claude Fable 5.2 or GPT-6.2 Astra, the two models that lead Neon in the Artificial Analysis Intelligence Index figures described in the source report.

The independent results are not uniformly favorable to Neon. On Terminal-Bench 4.0, Neon at xhigh scored 63.6%, below Carbon at high with 65.2%. On AA-Omniscience, Artificial Analysis reported a 14.4% hallucination rate for Neon, compared with 11.4% for Carbon. Google separately says Neon achieved its lowest fabrication rate yet in internal Antigravity testing; that is a company-reported result using a different measurement and is not independently verifiable from the supplied material.

““While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency.””

— Google

Where the Comparisons Fall Short

The supplied material does not provide enough detail to treat the different benchmark tables as directly interchangeable. Google’s 19-row evaluation and Artificial Analysis’s indexes use different test sets and comparison groups. Google’s table shows Neon ahead in 16 rows, but it omits the two models that rank above Neon on the independent Intelligence Index. The reason for that selection is not stated in the material.

It is also unclear how the reported benchmark costs and timings will translate to real-world workloads. Artificial Analysis figures depend on particular tasks and settings, while actual usage can differ in prompt length, output length, caching and pricing. The source gives Carbon’s temporary discount end date but does not establish whether that offer will be extended. Google’s internal fabrication-rate result cannot be checked against the independent hallucination measure cited here.

Neon trails Carbon on the reported Terminal-Bench 4.0 result despite its higher overall xhigh Intelligence Index score. The source does not establish why the models differ on that test, or how either result predicts performance across other coding tasks. Those gaps limit what can be concluded from a single aggregate score.

Testing Neon Against Real Workloads

The next practical step for prospective users is to test Neon at the effort setting and price they expect to use, rather than relying on the highest-setting score alone. Teams can compare quality, task completion time and token costs against their current model on representative work. The supplied source does not identify a scheduled follow-up evaluation or a new benchmark release.

Google’s announced availability includes the Gemini API in AI Studio, Gemini for business, and Google Antigravity and the Gemini app for paid Pro and Ultra subscribers. Google also said a cyber-defense version, Gemini 4 Neon for cyber, would reach eligible trusted defenders in its Fairwind Program over the next few days. The source gives no further access timetable or additional technical details for that version.

Pricing and independent measurements may change after launch. In particular, Carbon’s stated discount runs through November 30 in the source report, while Neon is a new model whose performance will need to be evaluated across more than the reported index tasks. Until then, the clearest confirmed comparison is limited: Neon posts a one-point top-setting index gain over Carbon in the cited data, with lower reported cost and time per index task at standard list prices.

Key Questions

What is Gemini 4 Neon?

Google released Neon on November 18, 2026, describing it as a model between Carbon and Hydrogen. Its model card says Neon is adapted from Carbon.

Did Neon score much higher than Carbon?

Not in the cited Artificial Analysis Intelligence Index comparison: Neon scored 61 at xhigh, while Carbon scored 60 at high. At the same high setting, Neon scored 60. These are specific test configurations, not a general measure of every task.

Is Neon cheaper to run?

Artificial Analysis reported a cost of $6.44 per index task for Neon at xhigh and $3.74 at high. Neon’s standard token prices are below Carbon’s standard prices, but Carbon was discounted through November 30, and real costs depend on usage and settings.

Does Google’s comparison table include the top independent-index models?

No. The supplied report says Google’s 19-row table compares Neon with Claude Opus 5.5 and GPT-6.2 Sol, not Claude Fable 5.2 or GPT-6.2 Astra, which rank ahead of Neon in the cited Artificial Analysis index.

What remains unknown about Neon?

The available material does not establish how benchmark results will carry over to individual workloads, why Google selected the competitors in its table, or how its internal fabrication-rate result compares with independent tests. More real-world and independently comparable measurements would help answer those questions.

Source: ThorstenMeyerAI.com

You May Also Like

Settle The Ground Rules Before Claude Meetings Joins Google Meet

Anthropic’s Claude Meetings beta adds a note-taking and document-making participant to Google Meet, with different privacy defaults for Team and Enterprise plans.

‘I don’t worry about a robot takeover’: AI expert Michael Wooldridge on big tech’s real dangers (and occasional blessings)

Oxford professor Michael Wooldridge discusses AI, big tech, and why he doesn’t fear a robot takeover, emphasizing the importance of understanding game theory.

The Critical 24-Hour Event That Illuminates AI Market Trends

Baidu’s open-source Unlimited-OCR and Mistral’s OCR 4 launched within 24 hours, illustrating contrasting strategies in AI market positioning.

Biohybrid Robots: Integrating Living Cells With Machines

Navigate the intriguing world of biohybrid robots, where living cells merge with machines, and uncover the groundbreaking potential that lies ahead.