OpenAI CEO Sam Altman Claims Internal AGI System Possible This Year
OpenAI CEO Sam Altman stated that the company has not yet achieved AGI, but expects to have an internal system by the end of this year that he would be willing to call AGI; this statement refers to internal capabilities and does not commit to a public AGI product.
OpenAI's charter defines AGI as a highly autonomous system that outperforms humans in most economically valuable work. This definition does not have a unified external testing threshold, and whether AGI is achieved mainly depends on OpenAI's own judgment of the system's autonomy, coverage, and performance.
OpenAI's Chief Research Officer Mark Chen estimates that the company is about 80% away from AGI. Co-founder Greg Brockman stated that looking back from two years later, the current period may be seen as the starting point for the birth of AGI; neither announced that existing public models have reached this standard.
Altman also expects that the next generation of key models may "invent new things" in a way that has actual value for the first time. If this judgment holds, it means that the model's capability goals are shifting from answering, programming, and task execution to being research-oriented systems capable of proposing and validating new scientific, engineering, or business discoveries; however, OpenAI has not disclosed the architecture, training scale, benchmark scores, or validation standards of this internal system.
The term "internal system" suggests that OpenAI may first deploy more powerful models in a controlled environment for research, code development, model improvement, or other internal workflows. This allows for testing reliability, abuse risks, costs, tool invocation capabilities, and safety controls before large-scale external release, but it does not indicate when it will provide equivalent capabilities to ChatGPT, APIs, or enterprise customers.
Market mechanisms suggest that AGI expectations will increase capital investment in computing power, data centers, electricity, chips, model development talent, and AI applications. OpenAI and its computing partners may benefit from stronger product and financing bargaining power brought by more powerful models; application companies relying on existing capability gaps face pressure from the native functionality of models sinking. The true determinant of commercial impact is not the "AGI" label, but whether the system can complete high-value tasks that can replace human labor in a controllable cost, stability, and compliant manner.
Source: Public Information
ABAB AI Insight
OpenAI's trajectory from GPT-3, ChatGPT to reasoning models and Agent tools has been pushing models from text generators to systems capable of invoking tools, writing code, and executing multi-step tasks. Its 2023 governance crisis shifted the debate from "when will we approach AGI, how to deploy safely, who has ultimate control" from a technical internal discussion to a core company governance issue; after the board restructuring, OpenAI continues to expand its research, computing power, and commercialization resource allocation. Altman has limited the goal for this year to "internal systems," reflecting the company's attempt to maintain a narrative of capability leadership while allowing for controlled deployment space.
On the capital path, "internal AGI" will first be used to accelerate OpenAI's own R&D: automating code generation, experimental design, data processing, model evaluation, and safety research, creating a positive feedback loop where models help develop the next generation of models. Subsequently, capabilities can be monetized through ChatGPT subscriptions, enterprise APIs, custom Agents, and infrastructure collaborations. The first beneficiaries will be the supply chain that can provide computing power, electricity, networks, data center construction, and professional evaluation services; the costs will fall on laboratories that need to continuously invest massive computing resources while addressing model safety and commercialization uncertainties.
Historically, DeepMind's AlphaGo and AlphaFold demonstrated that specialized systems can surpass human experts under clear objectives, but this does not equate to general intelligence; GPT-like models have expanded the coverage of language, code, and knowledge tasks to a broader range. For OpenAI's claimed AGI to be valid, it must bridge the gap between "excellent performance on benchmark tasks" and "long-term autonomous completion of most economic work in open environments." The industry is now transitioning from demonstrating model capabilities to reliably delivering Agents, where the true scarcity is not demonstration effects, but the auditable, controllable, and repeatable task success rates.
Essentially, this is about technological substitution. The AGI narrative shifts market focus from replacing individual job actions to replacing the continuous chain of discovery, planning, execution, review, and learning in knowledge work processes. The mechanism of change is the gradual combination of reasoning capabilities, tool invocation, long-term memory, and computing resources into a system that can operate autonomously; however, human jobs will not immediately disappear simply because an internal model is named AGI, as the speed of substitution is constrained by error costs, accountability, authority boundaries, data access, and organizational adoption cycles.
ABAB News · Cognitive Laws
Capability definitions can be completed internally, but value validation must occur externally.
Models can write answers, but systems produce results.
The bottleneck of technological substitution is not intelligence but responsibility.