← Founder Notes
Archive

Gpt-6 astra hit 72.6 on osworld 2.0 the same week google locks engineers into first-party models.…

Yethikrishna ROriginal on Threads

gpt-6 astra hit 72.6 on osworld 2.0 the same week google locks engineers into first-party models.

the agent wars just stopped being about benchmarks and started being about whose keyboard wins the next 100 million prs.

Context

OpenAI's system card of 3 September 2026 says GPT-6 Astra is the most capable model it has broadly deployed and that it is the first to reach the Critical level of cybersecurity capability under its Preparedness Framework.

How it compares

The OSWorld 2.0 score of 72.6 appears only in a secondary source, not in the OpenAI text I read. The claim that Google locks engineers into first-party models rests on one secondary report with an unresolved date. The 100 million PRs figure has no source I could find.

Watch next

  • The system card capability tables.

Sources

  1. GPT-6 Astra system card (OpenAI)deploymentsafety.openai.com
  2. OpenAI launches GPT-6 Astra benchmarks (The New Stack)thenewstack.io

Provenance

The note above is reproduced unedited from the original post, first published on Threads on 4 September 2026 at 09:03 IST. Sources are the papers and datasets the note draws on.

View the original post
Embed this note
<iframe src="https://founder.myndlabs.tech/notes/embed/gpt-6-astra-hit-72-6-on-osworld-Dc2bkbLjDXR" width="480" height="420" style="border:0;max-width:100%" loading="lazy" title="Gpt-6 astra hit 72.6 on osworld 2.0 the same week google locks engineers into first-party models.…"></iframe>

More notes