Ne oldu?
RTTNews reports that OpenAI introduced GPT-6 Astra, describing it as the company’s most intelligent and aligned model to date. The report attributes results, capabilities, rollout plans and pricing to OpenAI; none of those claims are independently confirmed in the supplied source.
RTTNews reports that GPT-6 Astra is designed for computer use, software engineering, cybersecurity, scientific research and professional work. According to the report, OpenAI says Astra scored 98 percent on FrontierMath Tier 4, 99.9 percent on ARC-AGI-3 and 100 percent on ExploitBench. RTTNews also reports scores of 64.6 percent on Terminal-Bench Science 0.1 and 57.9 percent on Terminal-Bench 4.0, compared with results attributed to Claude Fable 5.1 and GPT-5.6 Sol.
The report says Astra scored 59.3 percent on Agents’ Last Exam, ahead of the cited results for Claude Opus 5 and GPT-5.6 Sol, and that OpenAI highlighted a 96 percent score on GPQA Diamond. RTTNews says OpenAI presented Astra as capable of creating websites, analyzing data, producing presentations and spreadsheets, testing software and completing online forms. The report also says the launch version includes safeguards against advanced offensive cybersecurity tasks.
Kaynak ayrıntıları: rttnews.com ↗
Neden önemli?
If the reported results hold, Astra could materially affect how organizations use AI for software engineering, computer-based work, scientific research and cybersecurity. Its reported ability to complete multi-step tasks, combined with access through major cloud platforms, would make evaluation, oversight and cost comparisons important for prospective users.
The reported results span coding, scientific reasoning, computer-use and cybersecurity-related evaluations, making Astra’s significance broader than a routine model refresh. For organizations, the practical question is whether these claimed gains translate into reliable performance on real workflows such as software testing, data analysis and document production. The source provides no independent replication or methodology beyond the named tests and reported scores.
RTTNews reports that API pricing starts at $10 per million input tokens and $50 per million output tokens, with the tested configuration described as about 31 percent cheaper than the cited comparison on Terminal-Bench Science 0.1. If accurate, cost could influence model selection for high-volume workloads. The source does not establish total operating costs, performance outside the reported tests or whether the pricing applies to every Astra configuration.
İnteraktif Mekanizma: Aslında Nasıl Çalışıyor?
Bu gelişmenin arkasında yatan teknolojiyi etkileşimli olarak keşfedin.
What is a common training objective for an autoregressive language model?
Bundan sonra ne izlenecek?
The key questions are when broader access begins, which organizations receive the initial rollout, how the model performs in independent testing and how OpenAI limits high-risk cybersecurity use. The report does not provide a date for general availability or details of the launch safeguards.
RTTNews says Astra is initially rolling out to a limited group of organizations before becoming available to ChatGPT Plus, Pro, Business and Enterprise users. It also reports planned availability through the OpenAI API, Microsoft Azure and Amazon Bedrock. The report does not identify the initial organizations, specify regional restrictions, give a broader-release date or clarify whether all listed ChatGPT plans will receive identical capabilities.
Further scrutiny should focus on independent results, real-world reliability, failure rates in computer-use tasks and the scope of the cybersecurity safeguards. The source does not explain how offensive cyber capabilities are restricted, what monitoring is used or how users can verify the model’s claimed advantages.