← @bgSponge

GPT-6 Astra: A new generation of intelligence

Summary

OpenAI's GPT-6 Astra achieves state-of-the-art performance in science, coding, and cybersecurity by saturating benchmarks like ARC-AGI-3 with a 99.9% score.

  1. Astra advances scientific research workflows.

    In Terminal-Bench Science 0.1, Astra scored 64.6%, outperforming Claude Fable 5.1 while maintaining approximately 31% lower estimated API costs.

    Astra reaches a new high among the models compared at 64.6%

  2. Cybersecurity capabilities meet Critical thresholds.

    Astra can identify and develop zero-day exploits, scoring 99.2% on SRE-Bench tasks within four attempts to reverse engineer software binaries.

    meets the Critical threshold in cybersecurity

    Astra solved 88.0% of tasks in a single attempt

  3. Efficient reasoning reduces monitorability.

    Researchers found Astra's written reasoning harder to monitor than GPT-5.6 Sol because the model solves problems using fewer written steps.

    Astra’s written reasoning harder to monitor than GPT‑5.6 Sol’s

  4. Astra solves long-standing mathematical problems.

    The model helped establish a prime number bound of 186 and improved a large prime gap term that was unchanged for 80 years.

    Astra helped establish a stronger bound of 186

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →