Greg Brockman opened with “Welcome to the AGI era.” Skip the hype — the real story is that GPT-6 Astra scored 100% on ExploitBench, found zero-days in browsers and a cloud database during testing, and became the first model to hit the Critical cybersecurity threshold in OpenAI’s Preparedness Framework. OpenAI released it anyway, with the dangerous parts locked.
What it actually is
A frontier model in two tiers, Astra and Astra Pro, built to work inside your software rather than advise from the side: filling forms, updating records, driving a desktop like a human. Computer use is the headline capability — 72.6% on OSWorld 2.0, up from GPT-5.6 Sol’s 65.7%. Rolling out now to ChatGPT Plus, Pro, Business and Enterprise. HN gave it 938 points and 668 comments in a day.
The API and the gate
Standard access comes via the OpenAI API and AWS: $10 per million input tokens, $50 output, million-token context. The obvious use case is agents that operate real applications end to end. But the strongest cyber capability isn’t in the public API — that sits behind Daybreak, an approval-only program currently limited to security defenders doing vulnerability validation and malware analysis.
A lab treating its own model like export-controlled hardware — that, more than the AGI line, is the milestone.
You Might Also Like
- Two api Settings Tripled gpt 5 6 Sols arc agi 3 Score Openai Says the Harness not the Model was the Problem
- Openai Oauth Turns Your Chatgpt Subscription Into a Free Openai api but Should you use it
- Openai gpt Realtime 2 Translate Whisper Three Voice Models one api Several Startups Erased
- Openai Bets 250m on Chatgpt for Academic Researchers Free gpt 5 6 for 100000 Scientists
- Openai gpt 5 6 Luna 80 Price cut and Free User Access as Chatgpt Hits 1 Billion Weekly Users
Top AI Product 
Leave a comment