Sam Altman apologies after messy rollout of OpenAI's Astra
What's the story
OpenAI's latest AI model, Astra, has had a rocky start. The company's CEO Sam Altman publicly apologized for the "chaotic release," admitting that there were issues with the initial rollout. He assured users that OpenAI is working to resolve these problems and expand access to Astra soon. As with previous launches, Pro subscribers will be the first to gain access.
Model features
Astra is designed for complex computer-based tasks
Astra is OpenAI's latest attempt to push AI systems into complex computer-based tasks.
The model can handle multi-step tasks, create documents, spreadsheets, presentations, build websites, and test their functionality.
According to OpenAI, Astra performs better on computer-use tasks while completing them more quickly than its predecessor models.
Performance comparison
Model showed better performance in latency simulations
In latency simulations, Astra performed better than GPT-5.6 Sol on the offline subset of OSWorld 2.0, achieving stronger results in about 47% less time per task.
The company also claims that Astra is its best model for software engineering yet, outperforming GPT-5.6 Sol on DeepSWE v1.1 at an estimated API cost per task around 57% lower when comparing each model's highest-scoring configuration.
Access problems
Astra's launch affected by usage and capacity limits
Despite its impressive capabilities, Astra's launch has been marred by access issues.
OpenAI has been dealing with usage and capacity limits across its products.
Codex users were offered daily allocation resets to address access pressures, while the company has also reportedly slowed some frontier training activity while assessing network capacity.