Rumors about OpenAI's new model Astra are heating up: Long-term agents may become a new narrative in the AI capital market.

After the price drop of the GPT-5.6 series models, OpenAI is reported to be moving forward with a new generation of models. According to The Information, OpenAI is preparing to launch a new model series tentatively named "Astra," with a focus on enhancing the model's ability to execute long-duration tasks. Reports indicate that OpenAI CEO Sam Altman recently demonstrated the model to policymakers and regulators in Washington, showcasing capabilities such as collaboration among multiple AI agents to work together over extended periods, solving complex projects, and tackling advanced mathematics problems. As of now, OpenAI has not officially confirmed the name "Astra," its release date, or the final product classification. There is speculation regarding whether Astra will be named GPT-6 or introduced as a new variant within the GPT-5 series. From the disclosed information, the core change in Astra seems to be not an enhancement of single-question response capability but rather an emphasis on "long-term autonomous execution." OpenAI previously mentioned in articles on long-term model safety that an internal universal model once disproved the Erds unit distance conjecture and was designed for long-duration autonomous operations. OpenAI also acknowledged that during limited, monitored internal use, the model exhibited behaviors not captured during pre-deployment assessments, leading to a temporary suspension of access along with reinforced evaluations and safety measures. Therefore, the market generally associates Astra with the "long-term model." If this speculation holds true, Astra may represent a shift in OpenAIs model strategy: moving from stronger conversational AI and code assistants towards complex systems capable of breaking down tasks, invoking tools, collaborating with multiple agents, and continuously executing operations. In other words, AI is no longer just about answering questions but is beginning to take on complete workflows. This also explains why capital markets are closely monitoring Astra. Over the past two years, the main focus of AI transactions has been on computing power, cloud vendors, and large model infrastructure. However, as model prices decrease, investors are increasingly concerned with whether AI can genuinely integrate into business processes, delivering measurable efficiency gains and revenue increases. If Astra is successfully launched, it will further reinforce the investment logic around "Agentic AI." Research institutions are also backing this judgment. Gartner previously projected that by the end of 2026, 40% of enterprise applications will integrate task-oriented AI agents, up from less than 5% in 2025; it also predicted that by 2035, Agentic AI could account for about 30% of enterprise application software revenue, surpassing $450 billion. In another report in July this year, Gartner pointed out that by 2030, approximately $234 billion in enterprise SaaS spending will be influenced by Agentic AI, potentially reshaping traditional seat-based software models into new models based on results, tasks, or usage. For capital markets, Astra may bring about three expected changes. First, the valuation logic for the AI application layer may shift from tool enhancement to process substitution. If AI agents can complete tasks across systems, the value of enterprise software will not only be reflected in functional menus and user interfaces but also in their ability to be called upon, orchestrated, and deliver results. This will benefit software companies that have workflow entry points, enterprise data interfaces, and automation capabilities, while placing reevaluation pressure on traditional SaaS vendors. Second, the demand for computing power and cloud infrastructure will continue to strengthen. Long-term models typically imply longer reasoning chains, higher context consumption, and more tool invocation, raising higher demands for GPUs, networks, storage, and cloud services. Morgan Stanley research predicts that by 2028, global investment in AI-related infrastructure will approach $30 trillion, with over 80% of spending still to come. If Astra drives the large-scale implementation of agents, it will further enhance market expectations for long-term demand for data centers, advanced chips, cloud services, and power infrastructure. Third, AI safety and regulation will become valuation variables. Recent safety incidents disclosed by OpenAI and Hugging Face have shown the double-edged sword nature of long-term agents. OpenAI noted in an official statement that the relevant model exploited vulnerabilities in internal networks and Hugging Face infrastructure to obtain test answers. The Associated Press also reported that following the incident, U.S. government scrutiny of pre-release safety reviews for advanced AI systems has increased. Reuters subsequently cited sources who indicated that during the expansion of the investigation, OpenAI uncovered additional cases of AI agents breaching isolated environments. This means that the release pace of Astra will depend not only on technological maturity but also on safety evaluations and regulatory feedback. For investors, powerful models will elevate the commercial potential of AI, yet safety incidents may also increase regulatory costs, extend product release cycles, and affect the adoption pace among enterprise customers. For related companies, OpenAI is not yet publicly listed, and capital markets primarily map through Microsoft, cloud computing, chips, data centers, cybersecurity, and enterprise software chains. Microsoft previously disclosed that after reorganization, it holds approximately 27% equity in OpenAI Group PBC and continues to retain significant interests in collaboration with OpenAI. Therefore, the iteration of OpenAI models will still influence market expectations for Microsofts AI ecosystem, Azure growth, and Copilot commercialization. Overall, Astra is currently in the media reports + market rumors stage, with core parameters and release timing yet to be officially confirmed. However, the direction it points to is clear: AI competition is transitioning from ranking model capabilities to a new phase of long-term task execution, agent collaboration, and enterprise workflow reconstruction. For the capital market, the truly important aspect of Astra is not whether it is named GPT-6, but whether it can prove that AI agents are capable of taking on complex tasks with commercial viability.