Today for AI
HOT RADAR
llmHEAT 11.4°

AI CURATED HOT STORY · 10/15/2026

GPT-6-Astra Solves 98% of Routine Physics Research Problems at Under $3 Per Task

1 reports archived1 independent sourcesupdated 10/15/2026, 23:24:15
Synthesis & Latest Updates
1 Sources Cross-Validated

OpenAI's GPT-6-Astra achieved a 98% success rate on 50 routine theoretical physics problems within the Codex environment, costing under $3 per task. The benchmark targets 'turn-the-crank' research tasks—applying known methods to slightly modified questions—demonstrating high reliability and cost-efficiency for automating entry-level scientific computation.

LATEST/OpenAI's GPT-6-Astra achieved a 98% success rate on 50 routine theoretical physics problems within the Codex environment, costing under $3 per task. The benchmark targets 'turn-the-crank' research tasks—applying known methods to slightly modified questions—demonstrating high reliability and cost-efficiency for automating entry-level scientific computation.

HEAT TRENDHourly heat curve

2 fully observed hours
Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first
  1. The DecoderT2·78 pts
    • GPT-6-Astra solved 49 out of 50 physics benchmark problems without human hints, costing under $3 per problem.
    • The dataset focuses on 'turn-the-crank' routine research tasks (e.g., parameter tweaks, dimension conversions) rather than grand open challenges.
    • High success rates suggest AI can now effectively replace junior researchers for standard computational verification steps.
GPT-6-Astra Solves 98% of Routine Physics Research Problems at Under $3 Per Task | Today for AI