Astra's failure cause found: outdated Skill held it back, OpenAI fixed it and sent another reset.

Beating AI News Flash: OpenAI has fixed a batch of quality issues following Astra's launch. The most absurd one came from legacy Skills: some Skills written for previous models frequently misfired on Astra, and could even prevent the model from checking its own work.

Another issue came from a context management experiment, which caused Astra to stop prematurely and even reply to earlier messages. OpenAI has shut down the experiment, and an estimated 4,000 to 5,000 users were affected.

The team also took down a batch of misconfigured inference engines that were dragging down model quality. After the fixes, Astra should be able to follow the latest instructions more reliably and check its own work. OpenAI also announced another reset for users.