한국어EN
Evaluation & BenchmarksDavid Rein et al.

GPQA released

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

GPQA released — 448 questions; PhD-level experts score 65%, non-experts allowed 30 minutes of web search score 34%

Why it matters

Made difficulty that searching cannot solve its design principle, linking it to scalable oversight research

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2311.12022

Added in Week 35, 2026 · See that week

On the map — ⑯ Evaluation line

  1. Previous stopSWE-bench
  2. This recordGPQA
  3. Next stopGSM1k
See it on the map

Same week

Andrej Karpathy posts Intro to Large Language Models on YouTubeOfficialAnthropic releases Claude 2.1 with a 200K-token contextOfficialGreg Kamradt publishes the needle-in-a-haystack long-context recall stress testPressGAIA benchmarkPaper

Nearby on the same axis

SWE-bench releasedPaperHugging Face explains the MMLU score mismatch on the Open LLM LeaderboardOfficialGSM1k releasedPaperMT-Bench and LLM-as-a-judge publishedPaper