Overview
- This week OpenAI began a staged rollout of GPT-6 Astra to paid ChatGPT tiers and vetted partners while making the model available through the API, Microsoft Azure, and Amazon Bedrock.
- Astra is built to perform end-to-end computer tasks — filling forms, updating records, running QA, troubleshooting by watching screens, and producing finished documents — and it supports very large context windows for long workflows.
- OpenAI says Astra hit a 'Critical' level on its internal cybersecurity tests, including a 100% result on ExploitBench, and therefore has disabled exploit-generation features for the public while reserving them for vetted defenders through a Daybreak program.
- The release intensified industry debate after Nvidia CEO Jensen Huang called Astra evidence that 'AGI has arrived,' a claim that many researchers dispute and that has prompted rivals and cloud partners to urge multi-model strategies.
- Observers warn the model’s premium pricing, contested self-published benchmarks, and demo-driven use of artist assets create real commercial and legal risks and are likely to accelerate real-world testing, regulatory scrutiny, and diverse multi-vendor adoption.