Bindu Reddy 称 OpenAI 因对齐问题推迟发布新模型 Astra
Bindu Reddy 在 X 上爆料说 OpenAI 内部模型很强，但 Astra 卡在对齐上，可能下周就发。这是行业传闻，没有官方确认，不过它侧面说明了前沿模型的瓶颈已经从能力转向安全测试。关注 OpenAI 新模型发布节奏的人可以当线索看。
Airbnb 前 CTO、MyShell 相关投资人 Bindu Reddy 在 X 上发文称，OpenAI 内部已拥有能力很强的模型，但迟迟未发布下一版 Astra，卡点在于模型在真实环境中的负责任表现。她表示希望 OpenAI 已接近解决这些对齐难题，可能下周就能发布。该消息来自个人账号，OpenAI 官方并未确认 Astra 的存在或发布时间表，具体细节仍待验证。
如果说法属实，这条推文透露的核心信息是：前沿实验室的发布节奏越来越受对齐和安全测试而非算力或能力限制。文中没有给出参数量、基准名称或具体日期等可核实数据，读者需把它当作行业传闻对待，等 OpenAI 官方公告再下定论。
OpenAI is struggling with model alignment
They have extremely powerful models internally and are yet to release the next version of Astra
The entire problem is getting the model to behave responsibly in the wild
Hopefully they are close to the end of these struggles and will be able to drop something next week
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include OpenAI, Astra, alignment, Bindu Reddy.
Source attribution: curated from Bindu Reddy (@bindureddy). Cross-source corroboration: 11 independent sources have covered this development (including IT之家, Runway ML, elvis), confirming industry-wide signal strength.
