Unverified50% confidenceBenchmarkExact time
When GPT-4 was released in 2023, its bare-model score on the original version of ARC-AGI was only approximately 5%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/6/2026
First Seen
Valid until: 12/5/2026
Sources
Related Entities
Related Claims
UnverifiedGPT-5.6 Sol achieved a score of 92.5% on ARC-AGI-2.74% similarUnverifiedGPT-4 发布时在 MATH 数据集上的正确率为 42.5%71% similarUnverifiedEarly GPT-4 achieved a resolution rate of less than 2% on SWE-bench, while recent top models have surpassed 50%.71% similarUnverifiedAccording to information circulating in the community, GPT-6 achieved a high score on ARC-AGI-3 using a specialized reasoning harness, with the bare model capability reaching approximately 60%.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/865829API
curl https://kongchang.com/api/v1/knowledge/claims/865829MCP
get_claim(id=865829)