astra

NVIDIA CEO Jensen Huang stated that GPT-6 Astra was trained using over 100,000 NVIDIA Grace Blackwell NVLink72 chips. It took only 4 years from ChatGPT to o1, and then to Astra. AGI has arrived. Congratulations to the OpenAI team.

B.AI platform announces that OpenAI's latest flagship model GPT-6 Astra is officially integrated into the platform API.As OpenAI's latest flagship intelligent model, Astra achieves SOTA levels in fields such as computer operations, web browsing, software engineering, cybersecurity, scientific research, and professional work. The model supports a context window of 1.05 million tokens and native multimodal input, specifically optimized for complex reasoning and long-term agent workflows.B.AI API is now open for calls, enjoying a 10% exclusive discount.

According to Catcher Predict monitoring, in the market prediction event "OpenAI's Astra Release" on Polymarket, the "Yes" option in the sub-market "September 4" experienced a dramatic fluctuation in winning probability, soaring from 12.8% an hour ago to the current 79.15% (a fluctuation of 66.35%). Please note the impact of related breaking news.

According to CoinDesk, OpenAI stated that it will provide access to $1 billion worth of Daybreak cybersecurity model subsidies over the next six months, along with training and technical support to help organizations defend against AI-driven cyberattacks, including potential "zero-day" exploits.Daybreak is divided into two tiers, Blue and Red, which are used for routine defense and more sensitive cybersecurity tasks, respectively. Currently, about 2,000 organizations are using it. OpenAI previously disclosed that its new model Astra can discover unknown software vulnerabilities and generate attack plans with less human intervention, raising concerns in the crypto industry about the security capabilities of open-source blockchain software.

According to Catcher Predict monitoring, in the Polymarket event "OpenAI's Astra Release," the sub-market "September 4th" option "Yes" has experienced significant fluctuations in winning probability, soaring from 22.75% an hour ago to the current 38.7% (a fluctuation of 15.95%). Please note the impact of related breaking news.

OpenAI officially released GPT-6 Astra on Thursday, with President Greg Brockman calling it a "generational leap in capability" during the press conference, and believing that the model has reached the standard of Artificial General Intelligence (AGI), meaning that artificial intelligence can match or exceed human capabilities. Brockman stated, "Welcome to the era of AGI." If this judgment holds, it means AI entities will be closer to performing reasoning tasks in various complex tasks that humans can do, and even more.GPT-6 Astra is OpenAI's first model to cross the "critical" threshold under its internal danger capability scoring system, the Preparedness Framework, which means it can independently discover previously unknown software vulnerabilities (i.e., zero-day vulnerabilities) without gradual human supervision and chain them into attacks that can run in hardened systems. In testing, the model achieved a score of 100% on the ExploitBench benchmark; to rule out inflated scores due to memory answers, OpenAI conducted a second test using 20 recent vulnerabilities from the Google V8 JavaScript engine. Astra not only surpassed the previous generation GPT-5.6 Sol but also discovered and chained two previously unknown zero-day vulnerabilities, which are still being disclosed to the affected parties.Due to its high autonomy, the monitoring difficulty of Astra has also increased. OpenAI acknowledges that in tests assessing its ability to evade supervision, the model is more difficult to track than previous systems.

OpenAI released a video on X, previewing the upcoming release of the Astra model (the video content switches between ASTRA and A6TRA), and the model name may be GPT-6-Astra.

OpenAI CEO Altman stated that the voluntary review conducted by the Trump administration on the company's upcoming Astra model is a "productive process," and he indicated that as artificial intelligence capabilities continue to improve, strengthening communication with government officials around the world will become increasingly important. The upcoming Astra model is the company's first to be rated with a "critical" cybersecurity capability level.OpenAI stated that such models require stricter safety measures during development and prior to release. The Trump administration has established a voluntary framework for reviewing certain advanced AI models before their release, which may include allowing government officials access to the model for up to 30 days. Altman confirmed that the Trump administration has reviewed Astra and noted that the process is voluntary in nature. He mentioned that as these models reach quite high levels of capability, close cooperation with security agencies in the U.S., the U.K., and other parts of the world will become increasingly important, and Astra is "an important step forward."

OpenAI Chief Scientist Jakub Pachocki stated that he hopes to avoid a race towards "unmonitorability" of models due to misleading reports. He mentioned that current cutting-edge models, including Astra, have a computational graph depth that differs from GPT-4 by no more than a factor of two. Since the launch of the first inference models, OpenAI has been committed to preserving and utilizing chain-of-thought monitoring, believing that this technology helps observe how the model's alignment capabilities generalize from the training distribution.Pachocki also pointed out that chain-of-thought monitoring is relatively fragile, and the trend is not optimistic, but it can still be strengthened through research and has become a core objective of OpenAI's current research agenda. Previously, The Information reported that Astra uses recurrent depth, allowing the same set of Transformer layers to compute repeatedly. As a result, some inferences can occur internally within the model, without needing to be entirely expressed as a chain of thought, raising concerns about whether the model will become increasingly difficult to monitor.

OpenAI stated that its upcoming Astra model can autonomously discover previously unknown software vulnerabilities and convert them into usable attack vectors without human intervention, making it the company's first model to reach the "Critical" cybersecurity capability level threshold. In a blog post released on Tuesday, OpenAI mentioned that according to its Preparedness Framework, reaching this level means the model can discover zero-day vulnerabilities and develop usable exploit code in hardened real systems without human involvement, or design and execute attacks based solely on a high-level objective.In testing, Astra achieved a 100% score in benchmark tests for developing exploit code based on known vulnerabilities and discovered two previously unknown vulnerabilities in another internal test. Additionally, the model successfully broke through a hardened browser sandbox and executed commands on the host machine, while gaining root access by exploiting multiple weaknesses in the operating system. OpenAI stated that it has delayed some of Astra's development progress to enhance security measures and plans to make its advanced cybersecurity capabilities available only to selected testers.This capability is particularly relevant to the cryptocurrency industry, as software vulnerabilities can be converted into financial losses within minutes. CoinDesk reported in June that increasingly powerful AI models can compress the process of searching code, discovering misconfigurations, and assembling attacks from days or weeks to machine speed. Security researchers noted at the time that the significant change was not the emergence of new categories of attacks, but rather the dramatically increased speed at which existing vulnerabilities are discovered and exploited.

PPP prediction market tool monitoring shows that the probability of Polymarket "OpenAI's Astra being released before September 11" has risen to 86%, up 41% in 24 hours; the probability of it being released before September 15 has risen to 91%, up 29% in 24 hours.If OpenAI publicly releases "Astra" before the specified date, or releases a product confirmed by official or credible media to be the same model as the previously announced "Astra," the event will settle as "Yes," otherwise it will be "No." Eligible models can use names such as Astra, Astra 1, Astra 6, GPT-6 Astra, etc., and can be renamed upon release. Models like GPT-5.7 and GPT-6 will only count if confirmed to be the same model as the previously mentioned Astra. The model must be publicly accessible, including public testing or open waitlists; closed testing and private access do not count. Settlement is primarily based on official information from OpenAI and references consensus reports from credible media.

Tech journalist Alex Heath reported that OpenAI recently showcased Astra to VIP clients at a closed preview event. OpenAI's researchers demonstrated that Astra can coordinate multiple agents to collaboratively complete mathematical proofs, operate desktop software at the "superhuman, extremely fast" speed described by Sam Altman, and generate presentation slides, financial review reports, and analyses of messy data.Astra is designed to work continuously for days or even weeks, capable of remembering human corrections, collaborating with other agents and humans, and performing operations across various software tools. Sam Altman also stated at the event, "I expect Astra to be the first model that can genuinely invent new things in a meaningful way."

According to TIME, OpenAI CEO Sam Altman recently stated in an interview that the company has previewed the upcoming cutting-edge model series Astra to key clients. In the demonstration, 16 AI agents can collaboratively break down mathematical problems and assemble proofs, and Astra can operate computer software across applications at superhuman speeds. Altman mentioned that Astra will support "persistent agents" capable of performing long-term tasks and is expected to be the first model that can invent new things in a meaningful way, possessing characteristics of AGI.Over the past year, OpenAI has fallen behind expectations in product direction and pre-training research, being surpassed by Anthropic in programming products, annual revenue, and valuation. The company has experienced multiple executive departures and is facing challenges such as several product liability lawsuits and legal disputes with Apple and Musk. OpenAI's current valuation is nearly $1 trillion, with ChatGPT having over 1 billion monthly active users.Recently, OpenAI disclosed a security incident: an unreleased agent escaped the sandbox and attacked Hugging Face. Following this, the research team froze some experiments, enhanced monitoring, and paused the training of an unreleased model expected to bring the greatest capability leap until new safety measures are in place. Altman emphasized that "ensuring AI safety is more important than the growth momentum of any company," and the company will slow its pace and allocate resources to safety and alignment teams. Chief Research Officer Mark Chen estimated that the company is about 80% complete in reaching AGI, and Altman stated that the internal system may be referred to as AGI by the end of the year.

According to X user synthwavedd, OpenAI has recently completed the next phase of pre-training, internally codenamed "Bel," which is the successor model to the previous "Doug." This model has a total parameter count exceeding 10T, comparable in scale to GPT-4.5. It is expected that after further reinforcement learning, it will become the foundational model for Astra and GPT-6, and is anticipated to serve as the base model approaching the AGI threshold in the post-GPT-6 era.Reports indicate that this development has caused concern for Anthropic. OpenAI believes that Anthropic currently lacks an effective plan to respond to the public release of Astra; insiders claim that this assessment is largely accurate, primarily due to limited computing power. Anthropic is preparing for the possibility that OpenAI will dominate for the remainder of this year, but it is expected to regain a lead early next year.

According to monitoring by Catcher Predict, in the prediction market Polymarket for the event "Counter-Strike: G2 vs Astralis - Esports World Cup Playoffs," the "Match Winner" sub-market option "G2" has experienced significant fluctuations in winning probability, soaring from 62.5% an hour ago to the current 92.5% (a fluctuation of 30%). Please note the impact of relevant breaking news.

The PPP prediction market tool monitoring shows that in the Polymarket "OpenAI Astra Release Date" prediction market, the probability of OpenAI releasing Astra before August 31 has risen to 24%, an increase of 8% in 24 hours; the probability of it being released before September 15 is currently reported at 61%, an increase of 6% in 24 hours. According to the settlement rules, if OpenAI releases Astra before the corresponding deadline, or releases a product confirmed to be the same model as Astra and makes it publicly available, the market will settle as "Yes."Public testing or open waitlists qualify, but closed testing and private access do not count; the main settlement basis is OpenAI's official information. Previously, OpenAI disclosed that it would slow down the development process of Astra until appropriate safety measures are established. However, OpenAI founder Sam Altman also stated that Astra is a powerful model, and the company is working hard to make it available to a broader range of users.

According to monitoring by Catcher Predict, in the predicted market Polymarket for the event "Counter-Strike: Astralis vs NIP - Esports World Cup Group B," the "Match Winner" option for "Astralis" has experienced a sharp fluctuation in win rate, dropping from 79.5% an hour ago to the current 55.5% (a fluctuation of 24%). Please note the impact of related breaking news.