Google introduces Gemini 4 Neon: frontier intelligence with lower cost per task
Image generated by the Inferama Editorial Team
01

Launch and positioning within the Gemini 4 family

Google DeepMind introduced Gemini 4 Neon on November 18, 2026, consolidating its strategy of offering a range of models with different levels of capability and cost. This new model is strategically positioned between Gemini 4 Carbon, which is oriented toward long-duration and maximum-complexity projects, and Gemini 4 Hydrogen, designed for the rapid execution of everyday tasks.

The company describes Neon as a frontier-level model, capable of completing software engineering, knowledge work, and scientific workflow tasks with fast end-to-end completion. Its goal is not to replace the most powerful model in the family, but to offer a competitive alternative for demanding and frequent jobs that require a balance between reasoning depth and operational efficiency.

02

Technical capabilities and benchmark performance

According to evaluations published by Google, Gemini 4 Neon scores 61 points on the Artificial Analysis Intelligence Index at its most advanced reasoning level (xhigh), placing it third overall, behind larger and slower models. On the Vals Index, the model reaches second place with 70.9%, positioning itself on the cost-efficiency Pareto frontier.

In the field of programming, Google states that Neon leads the Artificial Analysis Coding Agent Index with 58.6 points, outperforming competitors such as Claude Opus 5.5 and GPT-6.2 Sol in certain agent tests. The model also stands out in scientific and mathematical reasoning, achieving the best results in its category on Terminal-Bench Science (70.5%), RiemannBench (84.0%), and HLE-Diamond (71.3%). It is important to note that these results come from tests selected and presented by the provider, so their generalization to all real-world usage scenarios requires independent validation.

Gemini 4 Neon performance in key benchmarks

Comparison of official scores in intelligence, programming, and scientific reasoning tests.

BenchmarkNeon ScorePosition or Context
Artificial Analysis Intelligence Index (xhigh)613rd place overall
Vals Index70.9%2nd place, Pareto frontier
Artificial Analysis Coding Agent Index58.61st place overall
Terminal-Bench Science70.5%Best in its category
RiemannBench84.0%Best in its category
03

Cost efficiency and execution speed

One of the main advantages of Gemini 4 Neon is its considerably lower cost per task compared to its predecessors. Google indicates that, at its default reasoning level (high), Neon reduces the cost per task by 32% compared to Carbon. Against the main competing models, the cost difference ranges between 50% and 80%, according to Vals Index and Agent Arena tests.

In terms of speed, the model is up to three times faster than Gemini 4 Carbon. This improvement is due to a higher output token speed (approximately 110 tokens per second compared to Carbon's 57) and lower token prices. Although Neon uses 15% more output tokens per task, the reduction in price per token and the higher speed compensate for this increase, resulting in an execution time approximately 40% lower.

04

Availability, pricing, and access modalities

Gemini 4 Neon is immediately available for several user profiles. Developers can access it through the paid API in Google AI Studio and Android Studio. Enterprises can integrate it via Gemini for Business, either through the API or from the model menu, as well as through Google Antigravity.

For consumers, the model is available in the Gemini app for subscribers to the Google AI Pro and Ultra plans. Users with free accounts do not have direct access to Neon, although they can use Gemini 4 Hydrogen through automatic model selection. The API price is 3 dollars per million input tokens and 15 dollars per million output tokens. Cached input tokens have a 95% discount, costing 0.15 dollars per million.

Gemini 4 Neon pricing and availability

Summary of rates and access channels confirmed by Google.

ModalityPrice (per million tokens)Access
Input tokens3.00 USDPaid API
Output tokens15.00 USDPaid API
Cached input tokens0.15 USDPaid API (95% discount)
Context window1,000,000 tokensUnified input/output
05

Security, cyberdefense, and current limitations

Google has developed a specific variant of Neon for cybersecurity, called Gemini 4 Neon for cyber. This version is designed to find, validate, and fix vulnerabilities in code. According to the announcement, this variant will reach trusted defenders within the Fairwind program in the coming days, without being immediately open to the general public.

Regarding general security, the model maintains Carbon's protections against misuse and prompt injection attacks. In Google's internal Google Antigravity tests, Neon recorded the lowest rate of fabricated information among the company's evaluated models. However, like any language model, it can still generate incorrect responses or hallucinate data, so its use in critical environments requires human supervision and result validation.

How to access Gemini 4 Neon

Steps to use the model according to the user profile.

  1. 01Developers: Log in to Google AI Studio or Android Studio and select the paid Gemini API.
  2. 02Enterprises: Configure Gemini for Business through the API or the model menu, or use Google Antigravity.
  3. 03Pro/Ultra Consumers: Open the Gemini app and select 'Neon' from the model dropdown menu.
  4. 04Cybersecurity Defenders: Request access to the Fairwind program through the Cloud account team to receive the specific variant.
06

Keep exploring

Sources consulted

03

Corrections and transparency

If you spot incorrect or outdated information, send us a correction with the page and source we should review.

Submit a correction