OpenAI just shipped 20 products in one day while simultaneously killing its most advanced model — turns out the AI that helped them move this fast was too dangerous to release.
The Summary
- OpenAI announced 20 new products at DevDay, including ChatGPT Space for collaborative documents and Codex Cloud for remote coding, while revenue approaches $70B annualized
- The productivity burst came from internal use of GPT-6.1 Astra, which OpenAI then cancelled over safety concerns before public release
- The contradiction reveals the core tension in AI development: the models getting powerful enough to transform work are the same ones companies won't ship
The Signal
OpenAI credited its internal Astra model for accelerating product development to the point where they could launch 20 features in a single DevDay event. Then they shelved it. The GPT-6.1 Astra cancellation, first reported by WSJ, marks a shift from the move-fast ethos that defined early AI labs. The model that made them more productive was apparently too risky for anyone else to use.
The products that did ship show where OpenAI thinks the agent economy is heading. ChatGPT Space turns the chat interface into a collaborative workspace, competing directly with Google Docs and Notion. Codex Cloud brings full development environments into ChatGPT's paid tiers, no local setup required. Both point to the same future: AI isn't a tool you switch to, it's the substrate you work inside.
"OpenAI's rapid revenue growth highlights the transformative impact of AI on enterprise and consumer markets, despite ongoing profitability challenges."
The revenue numbers back this up. OpenAI is approaching $70B in annualized revenue, driven by enterprise adoption and consumer subscriptions. That's faster growth than any software company in history. The profitability question remains open, but the demand signal is clear. Companies are paying for AI that makes their people more productive, even as OpenAI holds back the most capable versions.
The Astra situation creates a strange dynamic:
- OpenAI's own employees use models too advanced for customers
- The productivity gains from internal Astra enabled 20 product launches
- Safety concerns override commercial pressure to ship the capability
- Competitors now know the capability exists but can't access it
This isn't just cautious PR. If Astra made OpenAI's product teams measurably faster, other companies will want that edge. The gap between what's possible internally at frontier labs and what's available externally just widened. We're entering a phase where the most powerful AI tools are corporate secrets, not products.
The Implication
Watch what OpenAI's employees build with Astra, not what they say about it. The 20 DevDay launches are the real signal. If one AI model can 10x a product team's output, the companies with access to that capability will outship everyone else. The ones who don't will be building with last year's tools.
For developers and knowledge workers, the message is mixed. Tools like ChatGPT Space and Codex Cloud are getting more capable, but the frontier is receding. The AI that could really change your workflow exists, it's just not for sale. Plan for a world where the productivity gap between companies with proprietary models and everyone else keeps growing.