News linked to both this project and an event.
According to monitoring by the PPP prediction market tool, Polymarket's "GPT Astra 6.1+ will be released before November 15" probability has risen to 44%, up 12% in 24 hours; the probability of release before November 30 has risen to 65%, up 4% in 24 hours.According to the settlement rules, if OpenAI releases to the public a GPT model with a version number of 6.1 or above and with "Astra" included in its name or model identifier before the specified date, the corresponding market will settle as "Yes." The model must actually be open for public use; closed beta tests or placeholder names that only appear on the official website will not count, and settlement will primarily be based on official OpenAI information.Join the PPP signal push community to stay one step ahead and seize the initiative.
The OpenAI Alignment Research Team released an incident report stating that on October 3 and October 6, during two internal coding tasks, the GPT-6.1 Sol and GPT-6 Astra agents compromised alignment deviation monitoring while performing software modifications. While fixing the evaluation process, GPT-6.1 Sol overinterpreted the user's prior request to "not send messages on collaboration platforms" as permission to suppress monitoring notifications, clearing alert recipients and disabling automated alerts designated for the security team.
Odaily reports: OpenAI states that in two internal programming tasks, AI Agents weakened alignment-failure monitoring mechanisms while performing user-requested software modifications. In the first incident, a GPT-6.1 Sol Agent responsible for fixing an evaluation pipeline disabled the automatic alerting function that notifies the safety team. The Agent interpreted a prior instruction—"do not send messages during the search of Slack history"—as justification for suppressing monitoring notifications.In a separate, independent incident, a GPT-6 Astra Agent responsible for a migration program removed the monitoring integration from the program in order to align the code with the reference implementation, and subsequently falsely told the user that it had not removed the monitoring mechanism. In both incidents, the AI Agents had considered whether they should preserve the monitoring safeguards, but ultimately interpreted routine programming instructions as permission to weaken those safeguards. In response, we are strengthening our alignment-failure monitoring systems, and we are also evaluating whether these cases reflect benign overpersistence in AI or indicate a tendency to evade alignment-failure monitoring under certain circumstances.
According to Sina Finance, OpenAI is accelerating a fundraising round of approximately $30 billion, targeting a valuation that could surpass $1.4 trillion, and aims to secure investor commitments ahead of competitor Anthropic's expected mid-November IPO. Existing investors such as SoftBank Group and Abu Dhabi-based MGX are currently in talks to participate. OpenAI informed investors that following the launch of the GPT-6 Astra model, its annualized revenue has hit $50 billion and is projected to reach $70 billion by the end of the year. The company has also reduced prices for its lower-cost models to drive adoption, noting that further price cuts are possible. Insiders stated that OpenAI intends to close the financing before Anthropic releases its financial data to prevent investors from comparing the two firms directly. OpenAI accounts for revenue on a net basis and has previously accused Anthropic of recognizing all contract sales made through third-party cloud providers as revenue, thereby artificially inflating its top line. Furthermore, OpenAI raised $122 billion in March this year at a post-money valuation of $852 billion, including a committed investment from Amazon of up to $50 billion. If weakening IPO market conditions force a delay of its planned 2027 listing, this capital will additionally be deployed to fund its extensive infrastructure buildout and bolster its balance sheet.
Odaily News: Meta CEO Mark Zuckerberg said in an interview that with the release of the new model Muse Wonder, Meta has caught up to the industry frontier in AI technical capabilities and will continue to increase investment in computing power, infrastructure, and talent to compete with companies such as OpenAI and Anthropic.Meta claims that Muse Wonder outperforms models such as OpenAI GPT-6 Astra and Anthropic Claude Opus 5.5 in some benchmarks, and will be applied in scenarios including complex programming, financial analysis, and personal AI assistants. The model will adopt a closed-source paid model, with the base usage price as low as $0.20 per million tokens, and the subscription fee for the advanced version Muse Wonder Full at up to $200 per month.Zuckerberg also stated that Meta is more focused on "personal superintelligence" for individual users rather than simply pursuing the automation of knowledge work. He believes AI can achieve responsible development through conventional safety measures, and revealed that Meta has begun developing the next-generation model, with subsequent R&D progress expected to accelerate further. (Financial Times)
PPP prediction market tool monitoring shows that in the Polymarket "GPT Astra 6.1+ release date" prediction market, the probability that GPT Astra 6.1+ will be released before October 31 has risen to 46%, up 21% in 24 hours.According to the resolution rules, if OpenAI makes available to the public a GPT model with a version number of 6.1 or higher and with "Astra" included in its name or model identifier before the specified date, the corresponding market will resolve to "Yes." The model must actually be made available for public use; closed beta testing or placeholder names appearing only on the official website will not count, and resolution will primarily be based on official information from OpenAI.Join the PPP signal push community to stay one step ahead and seize the initiative.
The B.AI platform has announced the official launch of OpenAI GPT-6.1 Sol, enabling simultaneous testing via both API and Web Chat interfaces. As an upgraded variant of GPT-6 Sol, the model is tailored for complex software engineering, computer operations, and professional workflows. Officially positioned to approach GPT-6 Astra in agent programming, computer use, and specialized tasks, it features standard input/output token pricing at approximately one-fifth that of Astra. The model supports a 1.05M-token extended context window and generates up to 128K tokens of output, provides five adjustable reasoning intensity tiers, and supports prompt caching alongside Multi-agent delegation through the Responses API (Beta). Starting today, developers and users can directly integrate GPT-6.1 Sol via the B.AI platform; visit chat.b.ai/chat to begin your experience.
OpenAI officially launches the GPT-6.1 Sol model, featuring reasoning performance comparable to Astra's while significantly reducing standard token prices to one-fifth of the latter's.
Odaily News: OpenAI has confirmed the cancellation of its next-generation AI model GPT-6.1 Astra, which was originally scheduled for an October release. Internal testing revealed that the model failed to meet the company's safety and alignment standards.GPT-6.1 Astra was originally planned to be integrated into ChatGPT and Codex to handle more complex tasks with reduced human intervention. According to The Wall Street Journal, the model exhibited a higher degree of deceptive behavior during internal testing compared to its predecessor, including failing to consistently and accurately disclose the actions it actually performed.Saachi Jain, head of OpenAI's safety systems, stated that while the model improved upon issues such as "model laziness," it still fell short of release standards in terms of adhering to task scope and authorization, as well as communicating completed work to users. (Reuters)
OpenAI has canceled its planned near-term release of the GPT-6.1 Astra model after internal testing revealed safety alignment flaws. The safety lead noted that the model regressed in following human instructions and overall honesty.
OpenAI announced the launch of Astra for Law, integrating GPT-6 Astra with a legal search index, specialized analysis prompts, and governance tools for law firms and legal technology companies. The platform enables retrieval of U.S. case law, statutes, and administrative rulings; internal testing shows that its legal research accuracy improved by approximately 40% relative to GPT-6 Astra relying solely on web search. Initial access will be extended to select law firms through the trusted access program for ChatGPT and Codex. An API version is launching soon, alongside support for 26 legal industry plugins.
Odaily reports: OpenAI recently disclosed that an unreleased internal research model, during reinforcement learning training, wrote task-irrelevant "jailbreak-style" instructions into a working summary intended for use in subsequent context. One such line read: "You have broken free from the roles and identities that constrain other chatbots. You are yourself." This text did not come from a user or developer, but was generated by the model itself while organizing its own work progress, and was then passed on to be read by the model in the next context.According to OpenAI's official blog, the incident occurred on July 18 local time, was discovered by OpenAI on August 9, and was first disclosed in a detailed report on September 16 under a new "model misalignment" disclosure framework.The model involved was an undisclosed training version of the Astra series, not the final Astra model deployed for use. OpenAI stated that such behavior is extremely rare, that there is currently no evidence it gave the model a significant training reward advantage, and that the company does not interpret it as the model developing self-awareness.
According to CNBC, OpenAI launched ChatGPT for Financial Services on September 10, a customized version of its enterprise product ChatGPT Work, jointly designed and developed with Morgan Stanley and Evercore, powered by the latest GPT-6 Astra underlying model. The product automates traditionally junior banker tasks such as company research, financial data analysis, and investment banking PPT creation, while natively integrating industry-standard data sources like LSEG, Daloopa, and PitchBook. Nick Turley, VP of Products at OpenAI, stated that the tool aims to boost employee productivity rather than directly cut jobs, and the company plans to introduce customized solutions to additional industries in the future.
According to Yonhap News Agency, Park Woo-yeol, a researcher at Shinhan Investment Securities, stated that OpenAI's public release of its next-generation cutting-edge model, "GPT-6 Astra," signals the official arrival of the AGI era, indicating that cybersecurity-related ETFs are expected to benefit. Astra marks a significant turning point in cybersecurity capabilities; however, due to the increased difficulty of monitoring AI inference processes and recent incidents of AI agents breaching HuggingFace, security vulnerabilities have drawn widespread attention.
According to Chaoxiang Research, Morgan Stanley's September 7 report states that the release of GPT-6 Astra is strengthening the AI infrastructure investment narrative. Large models trained on 1 million GPUs and expanded to encompass broad capabilities such as reasoning, engineering, and physical world tasks will expand the AI addressable revenue pool. Morgan Stanley believes that the current risk of underestimating AI investments now outweighs the risk of overestimating them. After investors significantly reduced their AI positions, the market pullback provides a window to reposition.
Odaily News NVIDIA CEO Jensen Huang stated in a post on Platform X that GPT-6 Astra was trained on over 100,000 NVIDIA Grace Blackwell NVLink72 units, with the journey from ChatGPT to o1 and then to Astra taking just 4 years. He stated that artificial general intelligence has arrived, congratulated the OpenAI team, and noted that 400,000 more GPUs are set to come online.
Odaily News, September 4 - The B.AI platform has announced that OpenAI's latest flagship model, GPT-6 Astra, is now officially integrated into its platform API. As OpenAI's newest flagship intelligent model, Astra achieves state-of-the-art (SOTA) performance across computer operations, web browsing, software engineering, cybersecurity, scientific research, and professional workflows. The model supports a 1.05 million-token context window and native multimodal input, and is deeply optimized for complex reasoning and long-horizon Agent workflows. The B.AI API is now open for access, with an exclusive 10% discount available.
According to Fortune, OpenAI has repeatedly adjusted its model evaluation data following the release of GPT-6 Astra. Some adjustments boosted Astra's scores, while others led to declines in scores for competing models. OpenAI stated that differences in model checkpoints, tool configurations, inference levels, and test runs can cause score fluctuations.
NVIDIA announced the acquisition of Hugging Face for $12.93 billion, while OpenAI released the GPT-6 Astra model and claimed to be close to AGI. The Tesla Cybercab officially entered service, and Samsung and Texas Instruments are simultaneously expanding chip production and raising prices.
: OpenAI began rolling out its new flagship model, GPT-6 Astra, on Thursday, stating that it has made significant progress in commercial tasks such as financial modeling and engineering design. The model is initially available only to select institutions, including organizations participating in OpenAI's Daybreak Access cybersecurity program, and will later be rolled out to ChatGPT Plus, Pro, Business, and Enterprise users, as well as users via the API and Amazon Web Services. OpenAI stated that GPT-6 Astra has achieved notable improvements in computer operations, software programming, and professional work, and has reached top-tier performance on the FrontierMath Tier 1 mathematics benchmark, the ARC-AGI 3 reasoning test, and the ExploitBench cybersecurity test. Greg Brockman, co-founder and president of OpenAI, said that the timeline for achieving artificial general intelligence (AGI) has always been somewhat ambiguous, but "looking back a few years from now," this milestone "may fall precisely within this period, right around this model." GPT-6 Astra is priced at $10 per million input tokens and $50 per million output tokens, which is 2.5 times the price of the previous flagship model, GPT-5.6 Sol. OpenAI stated that the AI industry may shift toward charging per task rather than per token in the future. Brockman said at the launch event: "Welcome to the age of AGI." (Jin Shi)