While not a showstopper release, this release represents a strategic move. Google has known for a while that the main battle in AI is not won by the top-tier models, but rather by mid-tier models for developers and businesses. This week’s release of Gemini 4 Neon demonstrates this point.
Strategic Middle Model
With the release of Neon as part of the Gemini 4 range just 7 weeks after the last release, it would have been difficult to believe this would be possible even only a year ago. In terms of performance, this model sits between Gemini 4 Carbon (for high-performance needs) and Gemini 4 Hydrogen (for speed). Google states that this means that they have a good balance of performance, speed and cost. However, it is also an upgrade: Neon replaces both Argon and Gemini 3.1 Pro (notably, Argon was released on 1 st October and has already been replaced by Neon). Few tech products have such a short lifespan.
Key Features:
4 modes of reasoning. Neon has 4 modes of reasoning (low, medium, high, extra high), and 1 million token context window. This allows users to balance reasoning depth, speed and cost of operations.
High-performance coding. Google claims that this model is very high performance, and can complete end-to-end tasks quickly in software engineering, knowledge and scientific processes. Within Google’s development environment (Antigravity), Neon allows multiple sub-agents of Gemini 4 Hydrogen to operate in parallel. Security
In addition to coding, Neon has additional features for security professionals. The model retains the high level of protection seen in Carbon to prevent highly malicious use and prompt injection.
Key Figures
Some caution should be taken with the figures presented in this release. Different sources have different figures.
Multiple sources state that Neon is 3x faster and 60% cheaper than Carbon. However, a more balanced view comes from third party analysis firm, Artificial Analysis. They state that in comparison to Gemini 4 Carbon, Neon has only a small increase in intelligence, and that tasks using Neon are around 17% cheaper. Another source states that tasks are around 1/3 cheaper. The model has scored a 61 on the Artificial Analysis Intelligence Index, putting Google in third place overall in the AI industry.
This data can be combined. Although the price per token and speed can be compared, the total price of running a task depends on many factors. Therefore, although cheaper models may be expensive to run due to requiring more steps, this may reduce the benefit of cheaper prices. The most relevant figure for enterprise users is likely the 17% reduction. While this is only a slight reduction, it is real.
Availability
Neon is widely available from release. It is available for public use in Google’s API and Gemini Enterprise Agent Platform. Within the platform, Pro and Ultra users can use Neon within Antigravity and the Gemini application. Developers also have access to Neon in Google AI Studio. This wide availability indicates that Google want Neon in people’s daily workflows quickly to get ahead of competing models.
Comparisons to Competitors
Perhaps the most insightful articles compare Neon to competitors. Digital Trends notes that with Neon, Google are stating that they are able to compete with Claude Opus 5.5 and GPT-6.2 Sol at a lower price. This is a sensible move for Google. While they are not claiming to outperform the best models, they are claiming to provide most of the utility at a lower price.
This is a sensible move to make. While high level models garner media attention, the true economics of AI come down to the ability to perform large volumes of repeated tasks (such as code reviews, document analysis and automated pipelines). For these tasks, a model that provides nearly the same utility but at a clear discount is advantageous.
Risks of Speed
However, there are clear risks in the speed of this release. With 4 releases in only 7 weeks, it is difficult for developers to commit to using specific models. As the models are depreciating quickly, it raises questions around stability, cost of migration, and whether improvements in benchmarking truly reflect real-world improvements in models, or just optimisation for current leaderboards.
There is also a security concern. Although Neon contains security features for defenders, these could also be exploited by attackers. Google’s own concerns on this matter are clear in recent news, showing how close this line can be. They have raised concerns of AI being able to read a patch and create exploits.
Conclusion
Although not a breakthrough, this release is very strategically made. It shows that Google are willing and able to compete on efficiency, not just performance. Whether or not Neon remains competitive depends less on immediate benchmark results, and more on whether developers can trust models in such a rapidly changing series.

