Verified70% confidenceFactExact time
GPQA is a benchmark designed for PhD-level scientific question answering, with questions difficult enough that domain experts can only answer approximately 65% correctly.
4
Sources
70%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
UnverifiedGPQA is a PhD-level scientific question-answering benchmark designed by NYU researchers, with questions that avoid knowledge points answerable through simple searches.85% similarUnverifiedGPQA is a graduate-level scientific question-answering benchmark, and SWE-bench targets real-world software engineering tasks.79% similarUnverifiedPerplexity's inference service is built on GPUs.65% similarUnverifiedModern GPUs in inference scenarios are often bottlenecked by memory bandwidth rather than raw compute.65% similarUnverifiedA model with 64M parameters cannot handle complex reasoning, long contexts, or professional knowledge Q&A, and has a huge capability gap compared to mainstream large models such as GPT-4.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/554747API
curl https://kongchang.com/api/v1/knowledge/claims/554747MCP
get_claim(id=554747)