AI What Is AI Implementation for Businesses in Uzbekistan?
AI implementation is not adding a chatbot. It is embedding intelligent agents into real business workflows - lead qualif...
Gemini 4 Hydrogen is designed for everyday, high-volume work, while Gemini 4 Carbon is built for complex analysis and long-running tasks. The right approach for a business is not to choose one model and assign every task to it. Hydrogen should be the default model, with Carbon activated only when necessary.
This choice directly affects how Telegram inquiries are processed, how data from 1C or a CRM is analyzed, and how much an internal AI agent costs to run. But before comparing prices, two questions need answers: where is the data processed, and if the model makes a mistake, who makes the final decision?
On November 14, 2026, Spot.uz reported on the Carbon and Hydrogen models in the Gemini 4 family. Google had previously introduced Carbon on November 10 and Hydrogen on November 12.
Both belong to the same model family, but they were built for different types of work. Hydrogen is designed for fast, cost-efficient execution. Carbon is better suited to complex reasoning, coding, and long-running tasks.
| Criterion | Gemini 4 Hydrogen | Gemini 4 Carbon | Business decision |
|---|---|---|---|
| Primary purpose | Repetitive, high-volume work | Complex, multi-step analysis | Route tasks based on their type |
| Speed | Designed for fast responses | Prioritizes in-depth analysis | Use Hydrogen for customer queues |
| Price | $1.50 input, $7.50 output | $4 input, $20 output | Do not make Carbon the default |
| Typical work | Classification, short responses, data extraction | Planning, complex code, conflicting documents | Identify the task type first |
Prices are quoted per million tokens; Carbon rates take effect on December 1, 2026. According to Google Cloud pricing, Carbon costs about 2.67 times more than Hydrogen. That difference may be justified for some complex tasks. Sending simple questions to Carbon, however, is an unnecessary expense.
Gemini API is available in Uzbekistan. A company in Tashkent can obtain an API key and connect the model to a Telegram bot, website, CRM, or internal system. But using the service in Uzbekistan does not mean the data is processed there.
Google's data residency documentation indicates that global requests may be processed across different regions of Google's infrastructure. At present, no local computing region in Uzbekistan or approved on-premises deployment option has been specified for Hydrogen or Carbon.
That is why the entire data flow should be mapped before confidential information, such as a customer's passport or contract details, is sent to the model. Which fields leave the system? Which data is masked? What is written to the logs? These questions should be answered before choosing a model.
There is another practical limitation. A question received through Telegram is often about stock levels, prices, or outstanding balances. AI does not know this information on its own. It relies on up-to-date data from 1C, Odoo, or a CRM. Putting Carbon on top of incorrect inventory data only makes the mistake more expensive.

For an initial architecture, it makes sense to use Hydrogen as the default model. It can classify an inquiry by topic or extract a product name from a message. Carbon is also unnecessary for summarizing a conversation or preparing a response in Uzbek, Russian, or English.
Carbon should be called when the task requires comparing several sources, creating a long-term plan, reviewing complex code, or explaining contradictions between documents. Language is not a quality hierarchy. The test set should equally cover real requests in Uzbek, Russian, and English—the languages used by the business's customers.
The calculation is simple: monthly model cost = input tokens ÷ one million × input price + output tokens ÷ one million × output price. For example, assuming monthly usage of 100 million input tokens and 20 million output tokens, Hydrogen would cost $300, while Carbon would cost $800 at the new rate. That covers only the model. Integration, monitoring, and escalation to an operator are calculated separately.
The hidden cost often has little to do with token prices. Resending an unnecessarily long conversation history with every request drives up the budget. So does an agent repeating the same task or the absence of a queue for reviewing incorrect responses. Each call log should record the task type, selected model, category of data sent, response, and reason for escalation to an operator.
Neither model should independently make decisions that affect human rights and freedoms. Decisions such as denying credit, disciplining an employee, or issuing a medical conclusion must retain human approval and clear accountability. In simple processes governed by strict rules, AI may be unnecessary. Sometimes an API filter or a straightforward business rule is cheaper and more reliable.

At Celion, a project like this does not begin with the question, “Which model is more powerful?” First, tasks coming from Telegram, a website, or an internal dashboard are categorized. Next, the data fields exchanged with 1C, Odoo, and the CRM are defined. Hydrogen and Carbon are then tested on the same test set for cost, response time, error type, and escalation to an operator.
In a practical solution, a separate routing layer directs tasks between the models. Simple tasks go to Hydrogen, while complex ones go to Carbon. Risky or ambiguous cases are reviewed by an employee. Keeping rules and integrations separate from the model means the model can later be replaced without disrupting the business process.
The decision rule is clear: use Carbon only for tasks where its quality advantage justifies the additional cost and delay; handle the rest with Hydrogen or a precise business rule.
Celion's AI automation and systems integration service builds pilot solutions for securely testing Gemini models with Telegram, Odoo, 1C, or a CRM. The pilot routes tasks to the right model and tracks costs throughout the process.
Callback
Leave your details and we will contact you shortly.
Become a Partner
Leave your contact details and we will get back to you shortly.
Order Service
Leave your phone number and we will contact you shortly.