OpenAI ships GPT-6 Astra GA, trained on 100k+ Stargate GPUs; Brockman says AGI era has arrived

On September 14, 2026, OpenAI released GPT-6 Astra in general availability, moving Astra out of the limited-preview and gated-enterprise-access phase and into broad availability. OpenAI president Greg Brockman said during the launch event that the company has entered the AGI era, but stressed that AGI as a term is no longer tied to the contractual triggers in its agreement with Microsoft and should be read as a mission concept rather than a binding milestone. Brockman said he is leaving it to readers to decide whether Astra represents the real arrival of AGI, though he personally believes OpenAI has reached that stage. According to Aiden Clark, head of training at OpenAI, Astra is the largest training project the company has run, the first to use more than 100,000 GPUs at the Stargate site in Texas for pretraining, and the first OpenAI release in which earlier models heavily supervised the training process. In the initial rollout, Astra will first go to enterprise customers with access through the OpenAI Daybreak program, then roll out to ChatGPT Plus, Pro, Business, and Enterprise users over the coming days, and will be served through the OpenAI API as well as AWS. Pricing for the API is $10 per million input tokens and $50 per million output tokens, 2.5x the current promotional Sol pricing and in line with Anthropic's list price for Fable 5.1, well above Meta's standard $1.25 per million input and $4.25 per million output for Muse. Pro, Business, and Enterprise users will get access to GPT-6 Astra Pro, and eligible API customers can use Astra with zero data retention for workloads that require tighter compliance. OpenAI published the GPT-6 Astra safety white paper and Daybreak enterprise onboarding flow alongside the launch, noting Astra carries the same Critical-level internal risk rating as during preview.