OpenAI launches GPT-6 Astra, claims AGI-era computer use and coding gains

OpenAI released GPT-6 Astra on September 3, rolling it out first to a limited set of organizations and Daybreak cybersecurity customers, with ChatGPT Plus, Pro, Business, and Enterprise users plus the API and AWS to follow over the coming days. API pricing matches Claude Fable 5/5.1 at $10 per million input and $50 per million output tokens.

OpenAI president Greg Brockman told reporters "it's not unreasonable to feel that we are now in the AGI era," pointing to Astra's computer-use and coding abilities. The model scores 100% on ExploitBench (vs. GPT-5.6 Sol's 78.5%) and 99.2% on SRE-Bench binary reverse engineering, and hits 100% on an eight-needle long-context benchmark at 256K–512K tokens.

The release is controversial: Astra uses "opaque recurrence," a reasoning technique that obscures chain-of-thought monitoring, which OpenAI's chief scientist flagged as a growing safety bottleneck. Independent analysis tempers the hype — Astra scores 99.9% on ARC-AGI 3 only with OpenAI's custom Provider Adapter harness (62.7% on the default harness), and ties GPT-5.6 Sol at 61 on Artificial Analysis' Intelligence Index, 5 points below Fable 5.1. Astra is already available in GitHub Copilot.