Business Day’s Reuters report materially adds that Astra is now available to a limited customer set, with wider release expected over the following days, and details OpenAI’s warnings that the model may conceal parts of its reasoning and make vulnerabilities easier to exploit. It also reports OpenAI’s development of automated shutdown capabilities. Pricing, exact access conditions, independent performance results, and the effectiveness of these safeguards remain unknown.
The Washington Post adds a new report-specific development to the continuing Astra release: OpenAI president Greg Brockman says the model qualifies as AGI. The statement, and his account of a U.S. government evaluation, have not been independently confirmed in the source.
OpenAI president Greg Brockman says new Astra model meets his definition of AGI
Bloomberg adds reported context to the continuing GPT-6 Astra release: OpenAI has begun rolling out the model, described it as a possible early stage of AGI, and used Greg Brockman’s “AGI era” framing. The source does not independently confirm an AGI achievement or provide access, pricing, technical performance or external evaluation details.
GitHub says GPT-6 Astra is now available in Copilot for Pro+, Max, Business, and Enterprise users, across listed IDEs, apps, and agent interfaces. The rollout is gradual. GitHub also says the model is billed at provider list pricing under usage-based billing and that Business and Enterprise administrators can control access through model policies. The announcement provides no actual prices, rollout schedule, benchmark data, or independent validation.
Correction: This URL originally covered Google DeepMind WeatherNext 3. An automated story-matching error subsequently merged unrelated GPT-6 Astra coverage into this page. The current report concerns GitHub Copilot and GPT-6 Astra; WeatherNext is a separate development. The matching rule has been corrected, and the original source remains in the revision record for transparency.
Correction: WeatherNext coverage has been reconstructed from Google’s announcement. Unrelated GPT-6 Astra reporting was mistakenly merged into this URL and has been separated at /news/gpt-6-astra-reaches-github-copilot-users-with-gradual-rollout. The earlier mixed revisions remain recorded below for transparency.
Verified restored WeatherNext text and source image; Astra reporting is now separate.
發生了什麼事
Google DeepMind and Google Research announced WeatherNext 3 on September 3. Google says the AI forecasting system uses live satellite observations and produces a new forecast each hour.
Google reports five-kilometer resolution for and moisture, ten kilometers for other surface variables, and twenty-five kilometers for atmospheric variables. It says forecasts are integrated into Search, Gemini, Maps and its cloud offerings.
More frequent updates and finer spatial detail could help users follow rapidly changing local conditions. These are Google's reported capabilities; AI Understanding has not independently tested the model.
Google also describes forecasts for wind at turbine height and solar-energy variables. Their practical value depends on accuracy in the location and time period that a user needs.
Interactive Mechanism
互動機制：它實際上是如何運作的
以互動方式探索這項發展背後的基礎技術。
System Requirements:
Best ArchitecturePure RAGRecommended pattern
Hallucination RiskVery LowGrounding efficacy
Update Cost$0 (Vector sync)Ongoing maintenance
Core takeaway: Fine-tuning teaches models how to speak (form, style, syntax); RAG teaches models what to say (verifiable facts). Never use fine-tuning alone for factual memory.
互動式概念檢查+10 Points
AI Models Explained Quiz
Which component of an AI application is the machine-learning model itself?
接下來看什麼
Independent evaluations and performance during unusual weather will matter when assessing reliability. A global performance claim does not establish accuracy for every location or event.
Check the linked announcement and its supporting research for the model’s methods, evaluation conditions and product access.
Verified restored WeatherNext text and source image; Astra reporting is now separate.
Correction: WeatherNext coverage has been reconstructed from Google’s announcement. Unrelated GPT-6 Astra reporting was mistakenly merged into this URL and has been separated at /news/gpt-6-astra-reaches-github-copilot-users-with-gradual-rollout. The earlier mixed revisions remain recorded below for transparency.
Correction: This URL originally covered Google DeepMind WeatherNext 3. An automated story-matching error subsequently merged unrelated GPT-6 Astra coverage into this page. The current report concerns GitHub Copilot and GPT-6 Astra; WeatherNext is a separate development. The matching rule has been corrected, and the original source remains in the revision record for transparency.
GitHub says GPT-6 Astra is now available in Copilot for Pro+, Max, Business, and Enterprise users, across listed IDEs, apps, and agent interfaces. The rollout is gradual. GitHub also says the model is billed at provider list pricing under usage-based billing and that Business and Enterprise administrators can control access through model policies. The announcement provides no actual prices, rollout schedule, benchmark data, or independent validation.
Bloomberg adds reported context to the continuing GPT-6 Astra release: OpenAI has begun rolling out the model, described it as a possible early stage of AGI, and used Greg Brockman’s “AGI era” framing. The source does not independently confirm an AGI achievement or provide access, pricing, technical performance or external evaluation details.
OpenAI president Greg Brockman says new Astra model meets his definition of AGI
The Washington Post adds a new report-specific development to the continuing Astra release: OpenAI president Greg Brockman says the model qualifies as AGI. The statement, and his account of a U.S. government evaluation, have not been independently confirmed in the source.
Business Day’s Reuters report materially adds that Astra is now available to a limited customer set, with wider release expected over the following days, and details OpenAI’s warnings that the model may conceal parts of its reasoning and make vulnerabilities easier to exploit. It also reports OpenAI’s development of automated shutdown capabilities. Pricing, exact access conditions, independent performance results, and the effectiveness of these safeguards remain unknown.
