SIAgentNews.com
← All news
DevelopingGemini

Google updates Gemini 4 lineup with new models and pricing

Google has reorganized its Gemini 4 models, introducing Neon for Pro users and retiring Argon from the consumer app.

· First reported by Fello AI

14.8 minutesTime Neon takes to finish a task at High thinking

The short version

  • Carbon is Google's strongest model, while Hydrogen is the fastest and cheapest option.
  • Neon matches Carbon's performance at 'High' thinking but completes tasks significantly faster.
  • Argon has been removed from the Gemini app but remains available on the API.
  • Thinking levels (Low, High, Xhigh) can change a model's score more than the model name does.

Why it matters

Understanding these tiers helps you choose the right balance between speed and intelligence for your specific tasks. It also reveals that simply picking a premium model isn't enough; you must also check the 'thinking level' to get the best results.

The new Gemini 4 lineup

Google recently released four different Gemini 4 models over a five-week period in late 2026. These models are named after chemical elements—Argon, Carbon, Hydrogen, and Neon—though the names do not indicate their relative power or size. Instead, they represent different tiers of performance and speed for various user needs.

Carbon is positioned as the flagship model designed for the most difficult problems, such as long agentic coding runs. Hydrogen serves as the everyday model, offering high speed and low cost. Neon sits in the middle as a fast mid-tier model that balances these two extremes.

Argon was the first of the Gemini 4 models to launch but has since been moved to an API-only status. While it is still available for developers, it is no longer listed as an option for any personal Google AI plans. As Fello AI reported, the lineup is now structured to give users specific choices based on their subscription level.

Subscription and model access

Your specific Google AI plan determines which of these models you can access within the Gemini app. Free and Plus users get access to Hydrogen, which is also the only model available on the free API tier. For those on the Pro plan, Neon is the primary mid-tier option, while Carbon is reserved exclusively for Ultra subscribers.

The app uses a feature called Auto to manage these models for free users. Auto routes prompts between Hydrogen and a fast, lightweight model to determine the necessary thinking level for each request. This helps the system save on usage limits while trying to provide a helpful response for the user.

Users can manually check which model they are using by looking at the bottom of the chat box on the Gemini website. If a model has a lock icon next to it, it means that specific version requires a higher paid plan. Once a model is selected, users should also verify the thinking level, as this setting significantly impacts the output quality.

Performance and thinking levels

A key takeaway for users is that the 'thinking level' often matters more than the model name itself. For example, a model set to Low thinking can actually score lower than a cheaper model set to High thinking. Google notes that manually selected models default to Low, which may not provide the best results for complex tasks.

In testing, Neon at High thinking matches Carbon's score of 60 on the Artificial Analysis Intelligence Index. However, Neon is much faster, finishing a task in 14.8 minutes compared to Carbon's 36.6 minutes. This makes Neon a strong choice for most users who need high-quality analysis without the long wait times of the flagship model.

For the most accurate results, the guide suggests that users should set their thinking level to High if they are doing real work. While higher thinking levels use more of your plan's usage limit, they provide a significant jump in intelligence. Carbon loses the least amount of performance when turned down, but Hydrogen's score nearly doubles when moving from Low to High thinking.

API pricing and changes

The pricing for these models on the API is set to shift on December 1, 2026. Until then, Neon is the most expensive per-token model because Carbon and Argon are running at introductory rates. After the change, Carbon and Argon will double to $4 / $20 per million tokens, making Neon the cheaper of the two high-scoring models.

Hydrogen remains the most economical option, priced at $1.50 / $7.50 per million tokens. It is also the only model currently offered on the free tier of the API. This makes it the ideal choice for developers who need to process large amounts of data quickly without high overhead costs.

Argon continues to be sold on the API despite its removal from the consumer app. However, there is no official notice in the changelog regarding its deprecation. Unless you have a specific pipeline already pinned to Argon, the guide suggests moving to Neon, which scores higher and will eventually cost less per token.

What to watch for next

While there is significant buzz regarding a potential Gemini 4 Krypton model, it remains a rumor for now. No official benchmarks, model cards, or release dates have been provided by Google. Some social media posts suggest an internal checkpoint might perform better than Carbon, but this has not been confirmed by the company.

Google has warned that the Gemini lineup is subject to change without notice. This means the availability of models like Neon or Carbon could shift as Google updates its help pages and subscription tiers. Users should regularly check their specific plan's model picker to see what is currently included.

The current trend shows Google moving toward a more tiered approach where speed and intelligence are balanced by user choice. As more models are released, the distinction between 'fast' and 'smart' will likely become even more pronounced. For now, the best strategy is to pick the strongest model your plan allows and ensure the thinking level is set to High for important tasks.

Sources and links

First reported by Fello AI. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news