OpenAI's leadership is making a bold claim: the company will have an internal system that qualifies as artificial general intelligence (AGI) by the end of 2026. Chief Scientist Jakub Pachocki says the unreleased Astra model is the "Automated AI Research Intern" he aimed to build by September 2026, while CEO Sam Altman told TIME magazine that OpenAI will declare AGI achieved internally by December 2026. Mark Chen, another OpenAI leader, estimates the company is already 80% of the way there.

This timeline represents a significant acceleration in OpenAI's AGI predictions. Just nine months ago, the company's internal estimates were more cautious. Now, with the o-series reasoning models (o1 and o3) demonstrating increasingly autonomous research capabilities, OpenAI is confident enough to put a specific date on when it believes the threshold will be crossed.

What Does OpenAI Mean by "AGI" Exactly?

The term "artificial general intelligence" has been notoriously difficult to define, which is why most AI researchers avoid making specific timeline predictions. AGI typically refers to an AI system that can understand, learn, and apply knowledge across a wide range of tasks at or above human level. OpenAI's internal definition appears to center on autonomous research capability: a system that can independently identify problems, design experiments, and advance AI research without constant human guidance.

The Astra model, which Pachocki describes as an "Automated AI Research Intern," suggests OpenAI is measuring AGI progress by how well AI systems can perform the kind of knowledge work that currently requires highly trained researchers. If a system can reliably do what a junior AI researcher does, the logic goes, it has crossed into genuinely general intelligence territory.

How Are OpenAI's Reasoning Models Contributing to This Goal?

The o-series models, particularly o1 and o3, represent a fundamental shift in how OpenAI approaches AI capability. Unlike previous large language models (LLMs), which generate responses token-by-token in real time, reasoning models spend computational time "thinking through" problems before responding. This mimics human problem-solving more closely and enables the models to tackle complex research tasks that require planning and verification.

The progression from o1 to o3 shows measurable improvements in reasoning depth and autonomy. These models are being trained to handle the kind of open-ended research tasks that define AI research work: reading papers, identifying gaps, designing experiments, and iterating on solutions. OpenAI's confidence in reaching AGI by year-end appears directly tied to how well these reasoning models perform on research-specific benchmarks and real-world research tasks.

Steps to Understanding OpenAI's AGI Timeline and What It Means for You

• Internal vs. Public Declaration: OpenAI plans to declare AGI achieved "internally" first, meaning the company's leadership will make this determination before any public announcement. This gives the company time to assess safety implications and prepare communications.
• Reasoning Model Capability: The o1 and o3 models are the technical foundation for this claim. They represent a shift from pattern-matching to genuine problem-solving, which OpenAI sees as the key capability gap between current AI and AGI.
• Research Autonomy as the Metric: OpenAI is measuring progress by how well AI systems can conduct independent research work. An "Automated AI Research Intern" that can reliably contribute to AI development is OpenAI's operational definition of AGI.
• Timeline Compression: The acceleration from nine months ago to now suggests OpenAI has seen unexpected breakthroughs in reasoning model performance, likely from the o3 training process and deployment results.
• Industry Implications: If OpenAI reaches this milestone, it could reshape how other AI labs define and pursue AGI, potentially triggering a wave of similar claims or counterclaims from competitors like Anthropic, Google DeepMind, and others.

OpenAI has faced significant challenges in 2026, including key departures, rogue AI agents, and major lawsuits. In his TIME interview, Altman acknowledged these difficulties, stating that "we clearly had some missteps as a company". Despite these setbacks, the company's leadership remains confident in its technical trajectory toward AGI.

Altman

"There is not much time to act," Altman emphasized when discussing the urgency of AI development and safety coordination.

Sam Altman, CEO at OpenAI

The December 2026 target is less than four months away from the publication of this article. Whether OpenAI actually declares AGI achieved by that date will depend on how the company's internal evaluation criteria are applied and whether the o-series models continue to demonstrate the autonomous research capabilities Pachocki and Altman have described. The definition of AGI remains contested in the research community, so OpenAI's declaration, even if made internally, will likely spark debate about whether the threshold has truly been crossed.

What's clear is that OpenAI's reasoning models represent a meaningful shift in AI architecture and capability. Whether that shift constitutes AGI depends on how you define the term, but the company's confidence in a near-term declaration suggests the o-series has achieved something significant enough to warrant the label, at least in OpenAI's assessment.