OpenAI于2026年9月3日推出新型AI模型GPT-6 Astra,并宣称已"进入AGI时代"2。该公司将其描述为"能力的一代飞跃"1,CEO Sam Altman称之为"新能力水平"3。这是OpenAI首个达到其"关键网络安全能力阈值"的模型13,在网络安全、专业工作、软件工程、科学和计算机使用等领域表现突出1。OpenAI高管暗示这可能标志着通用人工智能真正被创建的时刻1,并表示该模型已通过正式审查3。
该模型首先向参与OpenAI Daybreak网络安全计划的企业集团推出,随后将在未来数天内向ChatGPT Plus、Pro、Business、Enterprise用户及OpenAI API和AWS推出3。在代码生成领域,GPT-6 Astra在Artificial Analysis代码生成代理指数中的token效率提升70%,幻觉率从92%大幅下降至51%6。在ARC-AGI-3基准测试中,该模型在标准环境下得分62.7%,在提供商适配器环境下达到99.9%,超越人类基准,且在96%的关卡上使用的动作数少于人类中位数7。
OpenAI CFO Sarah Friar表示,公司企业部门收入已超过消费业务,并预计"将在2027年成为上市公司",但如果业务表现良好可能更早上市3。该公司在6月向美国证券交易委员会秘密提交了招股说明书3。
OpenAI has unveiled GPT-6 Astra, declaring the arrival of the artificial general intelligence era 2. The company describes the model as a "generational leap in capabilities" 1, representing the culmination of years of research and significant investment 3. This marks the first OpenAI model to reach the company's "critical" internal cybersecurity capability threshold 3. An OpenAI executive suggested that if stakeholders look back in years to come, the moment AGI was truly created may be this very point in time with this model 1.
The rollout began with a limited group of enterprises participating in OpenAI's Daybreak cybersecurity program, with plans to expand to ChatGPT Plus, Pro, Business, and Enterprise users, as well as the OpenAI API and AWS, in the coming days 3. CEO Sam Altman characterized Astra as a "new level of capability" 3, while the model demonstrates exceptional performance across network security, professional work, software engineering, and scientific domains 1. The model has undergone formal review by the Trump administration, according to Altman 3.
In benchmark testing, GPT-6 Astra achieved a score of 62.7 percent on the ARC-AGI-3 standard environment at a cost of $26,000, and 99.9 percent in a provider adapter environment at $19,000, surpassing human baseline performance 7. The model showed a 70 percent improvement in token efficiency for code generation tasks 6, with hallucination rates dropping significantly from 92 percent to 51 percent at maximum effort 6. On the Artificial Analysis Coding Agent Index, it scored 67, matching performance with Claude Opus 5 and Fable 5, though at lower cost 6.
OpenAI's financial prospects are also shifting. CFO Sarah Friar informed employees that the company's enterprise division revenue has already exceeded consumer business revenue 3. Friar further stated that OpenAI "will become a public company in 2027," though it could go public earlier if business performs well 3. The company confidentially submitted a registration statement to the U.S. Securities and Exchange Commission in June 3.
评论
还没有评论,欢迎留下第一条。