OpenAI has officially unveiled its advanced flagship model, GPT-6 Astra — the culmination of years of intensive and rigorous research in pre-training, reinforcement learning, and alignment. This new milestone elevates the model to an unprecedented level of intelligence and operational capability.

Unrivaled Standing in Science and Technology:
Astra establishes new benchmarks across computer use, web browsing, software engineering, cybersecurity, empirical sciences, and professional tasks.

Shattering Global Benchmarks:
The model successfully saturated the complex FrontierMath Tier 4 benchmark with a score of 98%, solving historic and notoriously challenging mathematical problems. Furthermore, it achieved unprecedented landmark scores of 99.9% on ARC-AGI-3 and 100% on ExploitBench.

Unprecedented Speed and Efficiency:
In OSWorld 2.0 evaluations, Astra requires approximately 47% less execution time per task compared to its predecessor (GPT-5.6 Sol), multiplying its overall speed and operational throughput.

Safety and Operational Reliability:
The primary differentiator of GPT-6 Astra lies in its status as the safest and most aligned AI model ever developed. In rigorous stress tests evaluating edge cases and sensitive scenarios, while the previous generation (GPT-5.6 Sol) exceeded safety boundaries in 48% of instances, Astra registered a 0% boundary violation rate, demonstrating a profound comprehension of user intent.

Daily and Professional Workflows:
This advanced model seamlessly executes complex automated and routine operations with near-zero error margins:

Filling online forms and rapidly updating customer records within CRM systems.

Conducting deep online research, analyzing scientific datasets, and generating live websites and games directly from natural language prompts using the Sites capability in ChatGPT.

Organizing daily schedules, drafting email summaries and briefs, and editing and reformatting complex documents.