OpenAI ranks its path to AGI in five levels
OpenAI just shared an internal five-level framework to track its progress toward AGI, and executives say it is already closing in on Level 2. The ladder runs from basic conversational chatbots up to "Reasoners" capable of human doctorate-level problem-solving, all the way to systems that can do the work of an entire organization. It gives the clearest view yet of how the company defines human intelligence.
Everyone in tech loves debating when AGI will arrive, but nobody could ever agree on where the finish line even is, or how many miles we have left.
Now, OpenAI is pitching its own roadmap. According to a report from Bloomberg, an OpenAI spokesperson confirmed that the company shared an internal five-level tracking system with employees this week to gauge its progress toward artificial general intelligence. The company reportedly plans to share this framework with investors and the public as well.
OpenAI executives told staff that while the company is currently at level one, it is already knocking on the door of level two.
From conversational chatbots to full organizations
The hierarchy outlines a five-step ladder from today's text boxes to fully autonomous systems.
Level one is simply Conversational AI, covering standard natural language bots like today's ChatGPT.
Level two is dubbed Reasoners. OpenAI defines this tier as AI capable of basic problem-solving on par with a human holding a doctorate degree, specifically assuming that human has no access to external tools. Behind closed doors, executives even showcased a GPT-4 research project demonstrating human-like reasoning capabilities to employees, according to a person familiar with the meeting.
Level three is called Agents, describing autonomous AI that can carry out actions on behalf of a user over several days.
Level four, named Innovators, covers systems capable of aiding in new scientific or technological inventions. The ultimate tier, Level five, is called Organizations, meaning an AI system capable of outputting the total work of an entire organization.
OpenAI noted that this classification system will likely evolve as it gathers feedback from employees, board members, and investors.
Why it matters
Setting a five-stage milestone system is a smart corporate move, but it is also a reality check for the entire AI ecosystem.
By labeling today's systems as merely conversational and identifying reasoning as the immediate next step, OpenAI is effectively signaling where the next breakthrough will land. If you are waiting for a true AI agent that can reliably execute multi-day operational workflows without breaking, OpenAI is telling you that is a Level 3 ambition, and we are not even officially at Level 2 yet.
For enterprise buyers and developers, this means the near-term focus should remain on reasoning reliability rather than expecting turnkey autonomy. It also sets a clear benchmark to pressure rivals like Google and Anthropic to frame their own progress.
However, there is an obvious blind spot here. Defining Level 2 as doctorate-level reasoning without tools sounds impressive, but without public benchmarks, OpenAI remains the sole judge of its own rank.
It is also hard to ignore the timing. Back in May, OpenAI disbanded its Superalignment team, which was dedicated to mitigating existential AI risks. Pushing hard on an internal ladder toward superhuman organizational AI while dissolving core safety teams is bound to keep the safety debate louder than ever.
Sources (1)
- qz.com(opens in a new tab)Originalqz.com