Expired50% confidenceBenchmarkExact time
GPT models perform poorly when recall ability is tested in isolation, correctly recalling only about 50% of facts across all categories including spatial information, transcription content, and physical experiments, with open-source models performing even worse.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/17/2026
First Seen
Valid until: 10/15/2026(expired)
Sources
Related Claims
UnverifiedMultimodal large models have extremely low factual recall accuracy in video question answering, which is difficult to improve even with supervised fine-tuning.73% similarUnverifiedIn max mode, GPT-5.6's score is actually slightly lower than in ultra mode, confirming the overthinking phenomenon.69% similarUnverifiedGPT 5.2 scored approximately 77% on competition problems in frontier science benchmarks, but its performance dropped to approximately 25% on open-ended research tasks.69% similarUnverifiedExperiments conducted by the research team on models such as GPT-2 showed that after only 5–9 generations of recursive training, the quality of model outputs degrades significantly.69% similarUnverifiedResearch indicates that when the context exceeds a certain length, the model's recall accuracy for content in the middle portion may drop below 50%.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/543117API
curl https://kongchang.com/api/v1/knowledge/claims/543117MCP
get_claim(id=543117)