OpenAI has just unveiled GPT-6.1 Sol, a model designed for complex professional work and presented as almost as powerful as GPT-6 Astra, but at a lower cost. Behind this technical announcement lies a major economic challenge: making intelligent agents powerful and affordable enough for large-scale deployment in businesses. Combined with the new capabilities of the Agents API, GPT-6.1 Sol can search for information, use tools, and interact with computer interfaces. Here's what this evolution changes in concrete terms.
Key points to remember
- GPT-6.1 Sol is intended for development, computer use, and complex professional work.
- OpenAI claims performance close to GPT-6 Astra at a lower cost; this claim will need to be verified in real-world use.
- The model accepts up to 1.05 million context tokens and can generate up to 128,000 tokens.
- It can use web search, files, code, browsers and various connected tools.
- Lower costs can accelerate the deployment of specialized agents in companies.
- Human control, access security, and accountability remain essential.
Why GPT-6.1 Sol is an important announcement
In the race for artificial intelligence, performance alone is no longer enough. A company must also consider the cost of a task, its duration, the volume of data processed, and the reliability of the result.
An extremely powerful but overly expensive model might be suitable for a single exceptional mission but not cost-effective for thousands of daily operations. The GPT-6.1 Sol aims precisely to occupy this space between power and cost control.
OpenAI presents it as a model capable of approaching the performance of GPT-6 Astra on complex tasks, including software development, computer use, and professional work. However, this is not yet a universal truth; companies will need to compare it to Astra and other models using their own data and procedures.
What can GPT-6.1 Sol process?
GPT-6.1 Sol has a context window of 1.05 million tokens. This capacity allows it to receive very large volumes of text, code, or documents during a single mission.
For example, it could analyze a set of contracts, study the complete documentation of a software program, compare several technical reports, or monitor a project involving numerous sources.
The model can also use different tools through the OpenAI API.
- web search; ;
- search within files; ;
- image generation; ;
- code interpretation and execution; ;
- use of a computer; ;
- access to MCP servers and connected services; ;
- development tools and specialized skills.
In other words, the model is not only designed to produce text. It can become the engine of systems capable of searching, analyzing, and then acting.
How much does GPT-6.1 Sol cost?
The standard price announced by OpenAI is $2 per million tokens inflow and $10 per million tokens on exit. The cache of already processed data benefits from a lower rate. Surcharges may apply to very long contexts or certain processing methods.
These prices relate to the use of the model via the API and not to a consumer subscription to ChatGPT. They allow developers and businesses to integrate GPT-6.1 Sol into their own applications.
However, the price per token is not enough to determine the true cost. It is also necessary to measure the number of steps required, the use of paid tools, any rework needed after an error, and the human time spent on verification.
Why cost reductions change the game
When an agent works for several hours, uses different tools, or performs hundreds of daily tasks, every call to the model counts. A slightly less powerful but much less expensive model can then become more attractive than a premium one.
GPT-6.1 Sol could be used in several situations.
- analyze large sets of documents; ;
- prepare responses for a customer service department; ;
- monitor thousands of sources; ;
- produce reports and summaries; ;
- perform recurring checks; ;
- assisting developers; ;
- to operate several specialized agents.
Astra could be reserved for the most sensitive or difficult missions, while Sol would handle a larger volume of professional operations.
Agents capable of using websites and software
Alongside GPT-6.1 Sol, OpenAI is strengthening its infrastructure for agents. Its Agents API can provide a hosted environment, memory, tools, and sessions capable of lasting beyond a single request.
The computer's functionality allows an agent to interact with an interface, open a website, or follow certain steps in a process. However, the applications that use it must manage the necessary connections, permissions, and validations.
OpenAI also develops multi-agent systems. Rather than asking a single artificial intelligence to do everything, several agents can divide the work. One searches for information, another analyzes the results, and a third verifies or prepares the output.
A new work organization?
In a company, agents could function as specialized assistants. A sales agent would prepare a client file, a legal agent would identify clauses to review, and a communications agent would monitor market announcements.
This does not mean that the jobs in question will disappear. A professional mission is not simply about executing a list of steps. It involves judgment, negotiation, responsibility, and an understanding of the human context.
The most immediate impact should be on the organization of tasks. Professionals could spend less time gathering information and more time interpreting it, making decisions, or interacting with others.
What companies should check before adopting these agents
The deployment of an agent should not begin with the technology, but with a precise definition of the process to be automated.
The company must, in particular, establish several rules before its deployment.
- the information to which the agent can access; ;
- the actions he can perform alone; ;
- decisions that require human validation; ;
- the way in which its results will be verified; ;
- data storage and location; ;
- responsibility in case of error; ;
- the full cost of each automated process.
GPT-6.1 Sol supports data residency in the United States and the European Union according to OpenAI's specifications. However, some accelerated modes are not compatible with European residency. Organizations should therefore verify the actual configuration used, not just the model name.
The risks: error, autonomy and cybersecurity
An agent capable of using a browser or software can produce more value than a simple chatbot, but an error can also have more significant consequences.
Misinterpretation can lead to sending incorrect information, modifying the wrong file, or using data that should not have been accessible. Therefore, the rights granted must be limited to what is strictly necessary.
Companies will need activity logs, validation procedures, alerts, and mechanisms to interrupt an agent. Autonomy should not mean a lack of control.
Our reading
GPT-6.1 Sol is less visually striking than the new personal agents presented by OpenAI, but its economic impact could be considerable. To make agents widely available, OpenAI needs to ensure they are not only reliable enough but also affordable enough.
The battle for artificial intelligence is no longer solely about which model achieves the best result in a test. It now focuses on the cost of a complete mission, access to tools, security, business continuity, and the ability to integrate AI into existing processes.
Also read on Wazup-InTech, OpenAI unveils Dots and Space, when ChatGPT wants to work even in your absence.
