Before the end of this year, Sam Altman announces that OpenAI will have an internal system which he will call artificial general intelligence. The manager nevertheless admits that the company has not yet taken this step. This projection is mainly based on Astra, a new family of models capable of performing research tasks over several days. However, no autonomous test can certify that OpenAI is really at the threshold of AGI.
In brief
- Sam Altman plans an internal system comparable to the AGI before the end of 2026.
- Mark Chen estimates that OpenAI is 80% of the way toward this goal.
- Astra can already carry out certain tasks assigned to a novice researcher.
- An official AGI declaration will need to be verified by an independent panel.
Sam Altman plans internal AGI before the end of the year
According to statements by Sam Altman, OpenAI should achieve artificial general intelligence. The leader attests that the company is not there “not quite yet”but that it will be able to have at the end of this year an internal system that would qualify as an AGI.
He emphasized:
We haven’t quite reached AGI yet, but by the end of the year OpenAI will have a system internally that I would describe as that.
The selection of words provides a fundamental reservation. Altman confirms neither a public launch nor the official renewal of a universal technological threshold. He shares his own judgment of a model still in development and evaluated within OpenAI.
For Mark Chen, the company’s director of research, OpenAI is currently at “80% of the way” towards the AGI. However, no public protocol allows this calculation to be reproduced. Such a percentage therefore reveals the judgment of the laboratory managers, not the result of a standardized scientific indicator.
As for Greg Brockman, co-founder and president of OpenAI, he adopts an equally optimistic stance. For the latter, observers would consider in two years that the AGI emerged during the current period. These comments demonstrate that management no longer manages this technology as a distant objective.
The chosen definition remains essential. The AGI is then described in the official OpenAI charter as a set of independent systems capable of outperforming humans in the majority of activities of economic value. This approach therefore gives a central position to professional performance, without requiring the machine to reproduce all human faculties.
Astra already automates part of the work of researchers
It is on Astra that all the confidence of OpenAI, its next family of models, is based. During a presentation to clients, 16 agents managed to divide a research-level mathematical problem among themselves. In addition, they managed to coordinate their work before providing a demonstration proposal.
Astra would also have the ability to use various software on a computer and conduct missions over an extended period of time. These agents would work as virtual colleagues without waiting for a new instruction after each phase.
According to Jakub Pachocki, Scientific Director of OpenAI, the company has already achieved its internal goal related to automating the work of a beginning researcher. Thanks to an experimental idea, Astra would have the skills to write the necessary code within OpenAI systems, start the experiment and then present its results. The model could also accomplish some tasks that would occupy a researcher for almost a week.
This potential would help accelerate the development of next generations of models. An artificial intelligence capable of carrying out experiments could contribute to the improvement of its successor. Each progress made should therefore shorten the duration of the next cycle, a process usually referred to as recursive improvement.
However, the company remains more cautious in its official communication. In a plan released during the month of JuneOpenAI specified that a considerable part of its research work would be carried out simultaneously by humans and AI systems by March 2028. Reaching a first step this year therefore does not yet constitute the automation of all of its scientific activities.
The feats mentioned through on-chain data have not yet been the subject of a detailed technical publication. The company presented neither complete results, nor comparison with human researchers, nor evaluation led by an external structure. Astra remains an internal system whose capabilities are fundamentally described by its creators.
AGI recognition won’t depend solely on OpenAI
A difference exists between the system that Sam Altman might personally describe as AGI and an official company release. The theme does not have a unanimously recognized definition. This is why a model can excel in programming or research while remaining less excellent in new situations, physical work or decisions that require a deep understanding of the real world.
The partnerships between OpenAI and Microsoft enter another phase. Since their October 2025 revision, an official AGI communication by the company must be subject to careful review by a team of independent experts according to the Microsoft press release. Altman’s personal opinions could therefore not be sufficient to trigger the contractual obligations relating to this technology.
Such an audit has financial consequences. The partnership stipulates in particular that the sharing of gains between the two entities continues until the moment when the panel confirms the achievement of the AGI. Also, Microsoft also retains certain rights to OpenAI models and products until 2032, including systems designed after this stage.
This projection comes as OpenAi has slowed down certain tasks to consolidate its security measures. Astra’s exit would depend on further assessments, without a concise public deadline. A delay in marketing could not prevent the company from testing a more efficient system internally. However, he should limit the possibility of checking his abilities.
Ultimately, it should be noted that four elements will make it possible to judge this forecast by the end of this year: the dissemination of Astra’s evaluations, tests carried out by external researchers, a possible official declaration from OpenAI and the decision of the independent panel planned with Microsoft. The AGI announced by Sam Altman will fundamentally remain an internal qualification in their absence.
Maximize your Tremplin.io experience with our ‘Read to Earn’ program! For every article you read, earn points and access exclusive rewards. Sign up now and start earning benefits.
