OpenAI Previews 'Astra' Multi-Agent Model Family to US Senators, Undecided on GPT-6 Branding

Sam Altman demonstrated a new OpenAI model family, tentatively named Astra, to US senators in closed-door briefings this week, ahead of any public launch. The pitch is multi-agent collaboration: instead of one model carrying a task end to end, Astra splits complex problems into sub-tasks, hands them to separate agents working in parallel, and integrates the results, a design meant to get around the context and reasoning limits of a single model on long-running work.

• Altman met Senators Raphael Warnock and Bernie Moreno on July 29, with a session for Senate Intelligence Committee ranking member Mark Warner also on the schedule
• OpenAI has not decided whether to ship Astra as GPT-6, GPT-5.7, or a separate tier alongside its current Sol, Terra and Luna models
• Target use cases include extended coding projects, scientific research and business analysis that exceed what a single model can hold in context
• Astra is already in internal testing; no public release date has been set

Briefing regulators before a public launch is itself a signal: OpenAI is trying to get ahead of scrutiny on a model class built for autonomous, multi-step agent work, the same territory Anthropic and Microsoft are also pushing into. Where Astra lands on the coding leaderboard once it ships will matter more than the branding fight over its name.