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

🔍 Read the full analysis: Gemini 4 Neon: What Its Modest Gains Mean For AI Costs on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

  • Fast, free delivery on millions of items
  • Prime Video, Amazon Music and more included
  • Member-only deals all year
Start your free Prime trial Free trial for eligible customers · Cancel anytime
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 tested settings, alongside lower benchmark task costs and shorter completion times; results vary by effort level and test.

Google released Gemini 4 Neon on November 18, describing it as a model designed to balance capability and efficiency. In figures published by Artificial Analysis, Neon scored 61 on its Intelligence Index at the xhigh setting, compared with 60 for Gemini 4 Carbon at high; the measured index task cost was $6.44 versus $7.76, and completion time was 21.5 versus 37.1 minutes. The comparison uses different effort settings, so it is not a like-for-like score test.

At the same high effort setting, Artificial Analysis reports that Neon scores 60, matching Carbon at high, while costing $3.74 per index task. The published figures show Neon at 14.8 minutes per task at high. At xhigh, Neon gains one index point but the task cost rises to $6.44 and the time to 21.5 minutes. The results suggest that the effort setting can substantially affect the cost-capability trade-off.

Neon’s standard API rates are $3 per million input tokens and $15 per million output tokens, below Carbon’s standard $4 and $20 rates. However, Carbon was discounted to $2 and $10 through November 30, according to the source material, making it cheaper during that offer period. Artificial Analysis also reports that Neon generates output at about 110 tokens per second, compared with about 55 for Carbon, while using roughly 15% more output tokens per task. Cached input is listed at $0.15 per million tokens, a 95% discount from Neon’s standard input rate.

Google’s model card says Neon is adapted from Gemini 4 Carbon. It accepts text, images, audio and video, returns text, and has a one-million-token context window. Google lists four effort levels, with high as the default. The model is available through the Gemini API in AI Studio and several Google products, with access varying by subscription. Google says a cyber-defense version is being made available to eligible trusted defenders through its Fairwind Program.

At a glance
reportWhen: Released November 18, 2026; independent…
The developmentGoogle released Gemini 4 Neon, a model adapted from Carbon that posts a small independent index-score gain while lowering measured cost and time per 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 Each Extra Point

For teams running repeated coding, research or agent tasks, the practical question is not only which model scores highest but how much each completed task costs and how long it takes. Artificial Analysis’s figures make effort level a deployment decision: moving Neon from high to xhigh adds one index point but increases measured task cost from $3.74 to $6.44. That difference could matter at scale, though the benchmark’s task cost is not a universal price for every real-world workload.

Neon’s release also illustrates why model comparisons need more than a single headline score. The independent index places it behind GPT-6.2 Astra and Claude Fable 5.2 among model families, while the coding index puts Neon, Fable 5.2 and Claude Opus 5.5 within half a point of one another. In that coding test, the reported run cost for Neon is $6.86, versus $15.47 for Fable 5.2 and $12.61 for Opus 5.5. These are benchmark-specific results, not guarantees of equivalent savings for customer workloads.

Google says Neon brings “frontier-level intelligence with fast end-to-end task completion.” The outside measurements support a more bounded conclusion: Neon shows a modest score gain over Carbon in the cited comparison, alongside lower costs and faster completion under the reported conditions. Buyers still need to test their own tasks, including accuracy, output length and the cost of retries.

Products Worth Considering

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

As an affiliate, we earn on qualifying purchases.

Google’s Rapid Gemini 4 Releases

Neon is the fourth Gemini 4 model named in the source material in about seven weeks: Argon on September 30, Carbon on October 28, Hydrogen on November 12 and Neon on November 18. Google characterizes Carbon as suited to difficult, long-horizon work, Hydrogen as intended for fast everyday execution, and Neon as a balance between depth and efficiency. That positioning is Google’s description, not an independent evaluation.

The model card points readers to Carbon’s documentation for details on Neon’s architecture, training data and hardware, rather than presenting those details anew. Google lists a March 2026 knowledge cutoff, output of up to 64,000 tokens by default and up to one million tokens through background long decode. Independent benchmark results cited here were accessed November 19, 2026, and compare specific model versions and effort settings.

Google’s evaluation document reports Neon as best in 16 of 19 rows against Claude Opus 5.5 and GPT-6.2 Sol. Google’s table also shows those competitors ahead in three rows: Opus 5.5 on Terminal-bench 4.1 and FrontierCode 2.0, and GPT-6.2 Sol on Chartography. The table does not include Claude Fable 5.2 or GPT-6.2 Astra, which lead Neon in the cited Artificial Analysis family ranking. The separate evaluations use different test sets and should not be treated as one unified ranking.

“frontier-level intelligence with fast end-to-end task completion”

— Google

Limits of the Neon Comparisons

The cited Artificial Analysis comparison between Neon and Carbon uses xhigh for Neon and high for Carbon for the 61-versus-60 score and $6.44-versus-$7.76 cost figures. The same-setting high comparison gives both models a score of 60 and a Neon task cost of $3.74, but the source material does not provide a Carbon high-setting cost in that specific table. The figures are benchmark estimates, and the underlying task mix may not reflect a particular company’s usage.

Results are not uniformly better for Neon. On Terminal-Bench 4.0, Neon at xhigh scores 63.6%, below Carbon at high with 65.2%. Artificial Analysis reports a 14.4% hallucination rate for Neon on AA-Omniscience, compared with 11.4% for Carbon; the source cautions that this is distinct from Google’s internal testing. Google’s claim that Neon has its “lowest fabrication rate yet” in internal Antigravity testing cannot be independently checked from the information provided.

The material does not establish how Neon performs or costs across independent, real-world customer workloads, nor does it explain whether the listed Carbon discount will be extended. Google’s comparison table also omits the two models ranked ahead of Neon in the cited independent index. These gaps limit conclusions about which model offers the best overall value.

Testing Neon Beyond Benchmarks

Neon is available through Google’s listed API and consumer and business products, with some access tied to paid plans. Eligible Fairwind Program members were expected to receive access to the cyber-defense version over the next few days, according to the source material. The stated Carbon discount was due to run through November 30, 2026.

The next useful evidence will come from like-for-like independent tests that compare Neon and Carbon at the same effort setting, include leading competitors, and report quality alongside speed and cost. For organizations considering a switch, testing representative workloads at multiple effort levels can show whether Neon’s benchmark savings hold after accounting for longer outputs, retries and task-specific accuracy.

Key Questions

What is the main difference between Gemini 4 Neon and Carbon?

Google presents Neon as a balance of depth and efficiency, and its model card says Neon is adapted from Carbon. Artificial Analysis reports lower task costs and faster output for Neon in its measurements, but results depend on the effort setting and benchmark.

Is Neon more capable than Carbon?

In the cited Intelligence Index figures, Neon at xhigh scores 61 and Carbon at high scores 60. At high, Neon also scores 60. Because the one-point comparison uses different settings, it does not establish a like-for-like capability improvement.

How much does a Neon benchmark task cost?

Artificial Analysis reports costs of $0.75 at low, $1.90 at medium, $3.74 at high and $6.44 at xhigh per index task, using standard list prices. These are benchmark task estimates, not a fixed charge for every user task.

Is Neon cheaper than Carbon?

Neon’s standard rates are $3 per million input tokens and $15 per million output tokens, compared with Carbon’s standard $4 and $20. The source says Carbon was discounted to $2 and $10 through November 30, 2026, so it was temporarily cheaper on those rates during the offer.

Does Neon lead the independent model rankings?

No. In the cited Artificial Analysis Intelligence Index, Neon ranks behind GPT-6.2 Astra and Claude Fable 5.2. On the coding index, Neon, Fable 5.2 and Claude Opus 5.5 score within half a point, while reported task costs differ.

Source: ThorstenMeyerAI.com

You May Also Like

A Simpler Editing Workflow Starts With These 14 AI Video Tools

A buyer’s guide compares 14 video-editing listings, from PowerDirector and Adobe to VEGAS and VideoPad, with attention to versions, platforms and licenses.

Analyzing The Risks Of Narrowing The Astra Vs Fable Benchmark To Two Points

Analyzing the implications of reducing the Astra-Fable benchmark comparison to just two points, revealing potential misinterpretations and risks.

AI Agent Says Done, Database Says Otherwise: What To Check

Microsoft and Hugging Face released ThinkingBox, a benchmark checking whether AI agents leave business databases in the required state.

The Real Story Of OpenAI’s Jalapeño Chip In AI Innovation

OpenAI releases initial performance data for its custom Jalapeño inference chip, showing significant efficiency gains over NVIDIA hardware in AI workloads.