🔍 Read the full analysis: What To Know About Gemini 4 Neon’s Gains And Lower Bill on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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, presenting it as a model that balances capability and efficiency. Artificial Analysis measured a one-point index gain over Gemini 4 Carbon at the highest tested setting, alongside lower cost and shorter task times; the results vary by effort setting and benchmark.
Google released Gemini 4 Neon on November 18, pitching it as a balance of depth and efficiency. In figures published by Artificial Analysis, Neon scored 61 on the Intelligence Index at its xhigh setting, one point above Gemini 4 Carbon at high, while an index task cost $6.44 and took 21.5 minutes, compared with $7.76 and 37.1 minutes for Carbon at its standard price. The result suggests the clearest gain may be lower cost and faster completion, rather than a large jump in measured capability.
Artificial Analysis describes Neon as showing “a small Intelligence gain over Gemini 4 Carbon at a ~17% lower Cost per Task.” Its comparison uses Neon at xhigh against Carbon at high; these are different effort settings, so the scores are not a direct like-for-like test. At the same high setting, Neon scores 60, matching Carbon’s reported score at high, and costs $3.74 per index task. Neon at medium scores 55 for $1.90 per task, while low scores 44 for $0.75. Artificial Analysis reports task times of 14.8 minutes at high, 8.3 at medium and 3.7 at low.
Google lists Neon at $3 per million input tokens and $15 per million output tokens, compared with Carbon’s standard rates of $4 and $20. Carbon is discounted to $2 and $10 through November 30, according to the source material, which means Neon’s list-price advantage does not apply against that temporary offer. Artificial Analysis attributes the lower task cost and faster completion to Neon’s lower token prices and an output rate of about 110 tokens per second, roughly twice Carbon’s reported 55. Neon also uses about 15% more output tokens per task, offsetting some of the price difference.
The coding results point to a similar cost-versus-performance trade-off. On the Artificial Analysis Coding Agent Index, Neon scores 58.6, close to Claude Fable 5.2 at 58.5 and Claude Opus 5.5 at 58.2. Artificial Analysis reports Neon’s run at $6.86 per task, compared with $15.47 for Fable 5.2 and $12.61 for Opus 5.5. Those figures describe the tested runs and pricing assumptions; they do not establish costs for every customer workload.
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 Could Shape Model Choice
For teams that run models repeatedly, cost and completion time per task may matter more than a small change in an aggregate benchmark score. Neon’s reported results give developers a reason to test it for workloads where it can complete work faster without a meaningful drop in quality. The same data also shows a choice between capability and expense: raising Neon from high to xhigh adds one index point, but increases the measured task cost from $3.74 to $6.44.
The release does not settle which model is best for a particular job. Neon falls below Carbon on Terminal-Bench 4.0 in the reported comparison, scoring 63.6% at xhigh against Carbon’s 65.2% at high. On Artificial Analysis’s AA-Omniscience test, Neon’s reported hallucination rate is 14.4%, versus Carbon’s 11.4%. A lower bill could be useful, but buyers still need to compare accuracy, reliability and speed on their own tasks and at the effort settings they expect to use.
Products Worth Considering

Amazon Kindle Scribe (16GB) with Premium Pen, Tungsten

DJI Mini 5 Pro Fly More Combo with DJI RC 2, 1-Inch CMOS

LG UltraGear 27-inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDR True Black 400
As an affiliate, we earn on qualifying purchases.
A Fourth Gemini Model in Seven Weeks
Neon is the fourth Gemini 4 release in about seven weeks, following Argon on September 30, Carbon on October 28 and Hydrogen on November 12, according to the source material. Google describes Carbon as suited to hard, long-horizon projects and Hydrogen as aimed at fast everyday execution, with Neon positioned between them. Google’s model card says Neon is adapted from Gemini 4 Carbon; for details about architecture, training data and hardware, it directs readers to Carbon’s documentation.
Google says Neon accepts text, images, audio and video, returns text, and has a one-million-token context window. Its knowledge cutoff is March 2026. The model offers four effort levels, from low to xhigh, with high as the default. Google lists availability through the Gemini API in AI Studio, Gemini for business, and certain paid subscriptions to Google Antigravity and the Gemini app. A cyber-defense version is being offered to trusted defenders through Google’s Fairwind Program, with access for eligible members expected over the days following launch.
Google’s launch announcement calls Neon a model with “frontier-level intelligence with fast end-to-end task completion.” Its benchmark document reports Neon as the best result in 16 of 19 listed rows. In the other three, Google’s own table shows Claude Opus 5.5 ahead on Terminal-bench 4.1 and FrontierCode 2.0, and GPT-6.2 Sol ahead on Chartography. The table compares Neon with those two models, but does not include Claude Fable 5.2 or GPT-6.2 Astra, which rank ahead of Neon on the Artificial Analysis Intelligence Index.
““Frontier-level intelligence with fast end-to-end task completion.””
— Google DeepMind, launch post
Benchmark Coverage Leaves Gaps
Artificial Analysis’s figures are measurements from its index and model page, accessed November 19, 2026; they do not guarantee the same costs, times or scores on other prompts, infrastructure or usage patterns. The comparison also pairs Neon xhigh with Carbon high for the headline one-point gain, while the same-setting high comparison produces a tie. The source material does not provide a complete account of how each model was configured across every test, making settings and benchmark methodology relevant to interpreting individual differences.
Google’s table covers 19 rows but omits the two models that lead the Artificial Analysis Intelligence Index, so it cannot establish how Neon fares against the full set of top-ranked models. Google’s internal fabrication-rate claim is not independently verifiable from the information provided. It is also unclear how often the temporary Carbon discount will be extended, what actual customer bills will look like after usage patterns and other charges are included, and how Neon performs across a broad range of real-world tasks.
Pricing and Independent Tests Ahead
The immediate point to watch is Carbon’s listed discount through November 30: once it ends, the standard-price comparison with Neon may change for customers choosing between the two. Developers can also compare Neon’s low, medium, high and xhigh settings against their own tasks, since the reported data shows meaningful differences in both cost and score across those levels.
Further independent evaluations could clarify how Neon compares with the models omitted from Google’s table and whether its speed and cost advantages hold across different workloads. Google has not, in the supplied material, announced a date for additional benchmark results or a change to Neon’s pricing. The cyber-defense version is expected to reach eligible Fairwind Program members over the days after the announcement, but the precise rollout schedule is not specified.
Key Questions
What is Gemini 4 Neon?
It is a Google model released on November 18, 2026. Google’s model card says Neon is adapted from Gemini 4 Carbon and positions it between Carbon and Hydrogen in the Gemini 4 lineup.
Does Neon score higher than Carbon?
At the settings in Artificial Analysis’s headline comparison, Neon at xhigh scored 61 and Carbon at high scored 60. At the same high setting, the reported scores are both 60, so the size of the gain depends on which settings are compared.
How much does Neon cost?
Google lists $3 per million input tokens and $15 per million output tokens. Artificial Analysis measured a $3.74 cost per index task at high and $6.44 at xhigh; task costs depend on the tested workload and are not the same as a fixed charge for every user request.
Is Neon faster than Carbon?
In Artificial Analysis’s reported index tasks, Neon at xhigh took 21.5 minutes, compared with 37.1 minutes for Carbon at high. The source also reports output speeds of about 110 tokens per second for Neon and 55 for Carbon, but results can vary by task and configuration.
What is not yet clear about the results?
The supplied information does not show how Neon performs across all real-world workloads or provide a full independent comparison with the two models that lead the Artificial Analysis index. Google’s internal fabrication-rate claim is also not independently verified in the material.
Source: ThorstenMeyerAI.com
