OpenAI released GPT-6 Astra as its new flagship model. It leads on math, science, coding, cybersecurity, and computer use benchmarks. Astra scores 98% on FrontierMath Tier 4 and 99.9% on ARC-AGI-3. It completes tasks 47% faster than GPT-5.6 Sol on OSWorld 2.0. The model improves alignment, reducing unintended scope violations to 0% in tests. It rolls out first to select organizations. Plus, Pro, Business, and Enterprise users gain access over the coming days. API, Azure, and Bedrock access follows the same timeline. Astra refuses advanced cybersecurity tasks like exploit creation by default. Safeguards will loosen for defensive workflows in the coming weeks. It preserves context across long sessions in Codex via searchable notes. Astra handles complex professional workflows with better judgment and style adherence. It creates websites, apps, and games directly from prompts in ChatGPT Sites.
What changed
GPT-5.6 Sol was the previous frontier model with lower scores on math, coding, cybersecurity, and computer use benchmarks.
What it unlocks
Delegating complex, multi-step professional tasks with higher confidence in accuracy, speed, and alignment.
- 98% on FrontierMath Tier 4
- 99.9% on ARC-AGI-3
- 100% on ExploitBench
- 47% less time per task vs GPT-5.6 Sol
What you need to act on it
- ChatGPT Plus, Pro, Business, or Enterprise plan
- OpenAI API access
- Microsoft Azure or AWS Bedrock access
Sources