Unverified: GPT-6 Astra Reportedly Completes 279 of 280 Enterprise Agent Jobs
ryanshrout · x · 2026-09-06
Ryan Shrout shares first results from the Signal65 PINNACLE agentic benchmark claiming OpenAI's GPT-6 Astra posted remarkable enterprise agent numbers — unverified, treat with skepticism:
- At maximum reasoning effort, GPT-6 Astra finished 279 of 280 real multi-step enterprise jobs end to end.
- It fabricated nothing on unanswerable questions in the retrieval suite.
- Weighted errors were roughly half those of previous leaders Claude Fable 5.1 and Muse Spark 1.3.
- Cost per correct task at list API prices was also lower than the model it displaces.
PINNACLE's methodology: scenarios built from US Department of Labor occupation activities, procedurally regenerated sandboxes with runtime-generated answer keys (nothing to memorize), deterministic code-verified scoring with no model judges, and live agentic sessions. These claims come from a third-party post; the model names and results could not be verified — treat as an unconfirmed leak.
Related event: Leaked Benchmarks Claim GPT-6 Astra Aces Enterprise Tasks(2 posts)→
More from Models
- 105 Real Bugs Tested: GPT-6.1 Sol Over 10x Cheaper Than GPT-5.6 Sol — PawelHuryn · 2026-10-23
- Abliterated V4.1-Flash eval: abstract algebra drops 6pp but college math gains 5pp — teortaxesTex · 2026-10-23
- Claude users hit "model at capacity" errors hourly, canceling plans over it — Alex-Stack · 2026-10-23
- Kimi K3.1 launch imminent: 3T-param bet on RL seen as make-or-break — teortaxesTex · 2026-10-23
- Uncensored fine-tunes vanish from Hugging Face amid unexplained takedowns — umarmnaq · 2026-10-23
- Claude app menu leaks mystery model name 'Fable' as user hits limits — ezshine · 2026-10-23