OpenAI CEO Sam Altman says the company could have an internal system that he would describe as artificial general intelligence (AGI) by the end of 2026, putting one of the most consequential milestones in artificial intelligence potentially just months away. Altman told TIME that OpenAI has “not quite yet” reached AGI, but expects an internal system meeting his definition before the end of the year.
The prediction comes as OpenAI develops Astra, an upcoming family of models designed to perform extended research tasks and operate computer software autonomously. OpenAI chief research officer Mark Chen estimates that the company is about 80% of the way toward AGI, while co-founder Greg Brockman believes the current period could eventually be viewed as the point when AGI emerged. The claims remain internal assessments, however, and there is no universally accepted technical benchmark that determines when AGI has been achieved.
OpenAI Sets An End-2026 AGI Target
Altman’s latest comments represent one of his clearest timelines yet for AGI.
OpenAI’s own definition describes AGI as highly autonomous systems that outperform humans at most economically valuable work. That definition emphasizes the ability to perform a broad range of economically useful tasks rather than simply achieving human-level performance in one specialized area.
The distinction is important because current AI systems already outperform humans in certain narrow tasks. An AGI system, under OpenAI’s definition, would need to operate effectively across a much wider range of work with substantial autonomy.
OpenAI’s Current AGI Position
| Indicator | Latest Position |
|---|---|
| Current AGI status | Not reached yet |
| Sam Altman’s timeline | Internal system by end-2026 |
| Mark Chen’s assessment | ~80% of the way to AGI |
| Greg Brockman’s view | Current period could be remembered as AGI’s emergence |
| OpenAI’s definition | Highly autonomous systems outperforming humans at most economically valuable work |
| Public AGI benchmark | No universally accepted standard |
| Key upcoming model | Astra |
The wording “internal system” is also significant. Altman is not necessarily predicting a public announcement or consumer product explicitly labeled AGI by December 2026. He is describing a system OpenAI itself would consider to have crossed its internal threshold.
Astra Is At The Center Of OpenAI’s AGI Push
Astra is emerging as the key technology behind OpenAI’s confidence.
The upcoming model family is designed to operate as an automated research assistant capable of carrying out complex tasks over extended periods. OpenAI researchers told TIME that Astra can take an experimental idea, write code, run experiments in OpenAI’s codebase and report the results. It can also work through research papers and perform tasks that previously could have taken a human researcher about a week.
During a customer demonstration, 16 AI agents reportedly divided a research-level mathematics problem into smaller tasks, coordinated their work and combined the results into a proposed proof. Astra was also demonstrated navigating desktop applications and creating or editing work across different software programs.
What Astra Is Designed To Do
| Capability | Potential Impact |
|---|---|
| Code generation | Automates parts of software development |
| Experiment execution | Allows AI to conduct research tasks |
| Research-paper analysis | Accelerates information synthesis |
| Multi-agent coordination | Enables complex problems to be divided into subtasks |
| Desktop operation | Allows AI to work across applications |
| Persistent agents | Supports longer-running autonomous tasks |
| New-knowledge generation | Could accelerate scientific and technical discovery |
The ability to perform tasks continuously rather than simply respond to individual prompts is particularly important to OpenAI’s AGI ambitions.
From Chatbots To Persistent AI Workers
The development of persistent agents represents a broader change in how OpenAI views AI systems.
Traditional chatbots generally operate one interaction at a time. A user asks a question, receives an answer and then decides what to do next.
An agentic system can instead be assigned an objective and continue working toward it.
User Goal
↓
AI Breaks Task Into Steps
↓
AI Uses Software And Tools
↓
AI Runs Experiments / Performs Research
↓
AI Reviews Results
↓
AI Continues Until Task Is Completed
This type of autonomy is closer to the capability implied by OpenAI’s AGI definition.
It also explains why the company is increasingly focused on systems that can function as virtual colleagues rather than simply conversational assistants.
OpenAI Says AI Could Begin Accelerating AI Research
One of the most important implications of Astra is its potential use in AI research itself.
If AI systems can conduct experiments, write research code, identify problems and propose new approaches, they could potentially accelerate the development of subsequent AI systems.
OpenAI has previously described an automated AI researcher as a major milestone. In April, Altman said the company expected extremely capable models soon and discussed the potential for automated AI research to accelerate the pace of AI development.
This creates a possible feedback loop:
AI Researcher
→ Faster AI Experiments
→ Better AI Models
→ More Capable AI Researcher
→ Faster Research
The potential for this loop is one reason AGI timelines have become increasingly important for governments, businesses and investors.
However, the existence and speed of such a feedback loop remain uncertain.
AGI Still Has No Universal Finish Line
One of the biggest complications surrounding Altman’s prediction is that AGI does not have a universally accepted technical definition.
OpenAI’s definition is based heavily on economic performance: systems that can outperform humans at most economically valuable work. Other researchers may require broader capabilities, including robust reasoning, transfer learning, long-term planning, real-world understanding and reliable performance across unfamiliar environments.
This means two organizations could evaluate the same AI system and reach different conclusions about whether it qualifies as AGI.
Why AGI Definitions Matter
| Question | Why It Matters |
|---|---|
| What counts as “general”? | Determines breadth of required capabilities |
| What counts as “autonomous”? | Determines how much human supervision is allowed |
| Which jobs count as economically valuable? | Changes the performance threshold |
| How reliable must the system be? | Separates demonstrations from dependable deployment |
| Who evaluates AGI? | Determines whether the claim is independently credible |
| Does it need scientific discovery? | Raises the capability bar significantly |
Consequently, an OpenAI declaration that it has achieved AGI would represent an important company milestone, but it would not necessarily end the broader scientific debate over whether AGI had actually been achieved.
OpenAI’s Leadership Is Increasingly Confident
Altman’s forecast is not an isolated statement from OpenAI leadership.
Chief research officer Mark Chen reportedly believes the company is approximately 80% of the way toward AGI. Greg Brockman has expressed an even more optimistic view, suggesting that people looking back from the future may regard the current period as the moment AGI was created.
The combination of these views suggests that OpenAI’s internal expectations have shifted significantly closer to the present.
At the same time, the company’s leadership has acknowledged uncertainty around the pace of progress.
OpenAI’s official materials continue to emphasize that the timeline to AGI is uncertain, even as the company describes increasingly capable systems as coming into view.
OpenAI Is Rebuilding Around Agentic AI
The AGI push is occurring alongside a major change in OpenAI’s product strategy.
The company is increasingly combining models with tools, coding systems, persistent agents and workplace applications. One internal effort called “The Merge” combines Codex with ChatGPT, while OpenAI is developing products intended to bring agentic capabilities to a much wider consumer and business audience.
The shift can be represented as:
ChatGPT
→ AI Assistant
→ AI Agent
→ Persistent AI Worker
→ Automated Researcher
→ Potential AGI
The commercial importance is substantial because an AI system capable of completing tasks rather than merely answering questions could command much greater economic value.
A Potential AGI Would Change The AI Industry
If OpenAI actually reaches its internally defined AGI threshold by the end of 2026, the implications could extend well beyond the company.
Software development, scientific research, customer service, financial analysis, consulting, education and other knowledge-intensive industries could face rapid changes as increasingly autonomous systems become capable of performing larger portions of human workflows.
The biggest potential impact may come from scientific research.
AI systems capable of conducting experiments and generating new hypotheses could accelerate work in areas such as materials science, medicine, engineering and computer science.
But the transition would also raise difficult questions around employment, intellectual property, safety, cybersecurity, concentration of technological power and regulatory oversight.
OpenAI’s AGI Claim Would Face Scrutiny
A declaration of AGI would likely receive significant attention from governments, competitors and independent researchers.
The key question would not simply be whether OpenAI calls a system AGI, but whether the system demonstrates broad, reliable and repeatable capabilities consistent with the company’s definition.
Independent evaluations could therefore become important.
A system that performs brilliantly in demonstrations but requires extensive human intervention would represent something different from an AI that can independently perform most economically valuable cognitive work.
The Bigger Picture
Sam Altman’s end-2026 prediction represents a major acceleration in the public timeline for AGI. OpenAI is no longer describing general intelligence as a distant theoretical objective; its leadership believes increasingly autonomous systems such as Astra could bring the company close to its internally defined threshold within months.
The claim should nevertheless be treated as a forecast rather than a confirmed technological milestone. AGI lacks a universally accepted definition, Astra’s full capabilities have not yet been independently evaluated at scale, and the difference between impressive AI demonstrations and reliable autonomous economic performance remains significant.
Looking Ahead
The remainder of 2026 will provide important evidence for assessing OpenAI’s forecast. The performance of Astra, the extent to which it can independently conduct research and the reliability of its persistent agents will be closely watched. An eventual internal declaration of AGI would also raise questions about how the company defines the milestone and what evidence it uses to support the conclusion.
If OpenAI does reach its own AGI threshold this year, the significance would extend far beyond a product launch. It could mark the beginning of a period in which AI systems increasingly participate in research, software development and other high-value cognitive work. Whether that transition happens in 2026, later or under a different definition of AGI, the rapid shift toward autonomous AI agents is already reshaping the industry’s expectations.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
