Astra nears benchmark saturation, scoring 98%+ on 10 of 37

Evaluations show Google's Astra scores at or above 98% on 10 of 37 benchmarks, versus only 4 for GPT-5.5, indicating the suite is nearing saturation. This makes reliable time-horizon estimates difficult despite new estimation methods using synthetic long-horizon tasks.

2026-09-09 ~ 2026-09-09 · 2 related posts

• Astra scores 98%+ on 10 of 37 benchmarks, nearly saturating eval suite — dswg97 · 2026-09-09
• Astra hits ≥98% on 10 of 37 benchmarks, near-saturating the time-horizon eval suite — dswg97 · 2026-09-09