GPT-5.6 Sol scores 7.8% on ARC-AGI-3, first AI to complete a benchmark game task

OpenAI's GPT-5.6 Sol model scored 7.8% on the ARC-AGI-3 benchmark and became the first verified frontier model ever to complete one of the games included in the test, according to results published by the ARC Prize foundation on 9 July. Previously, no model had surpassed roughly 2% on this benchmark, making the result a significant milestone. ARC-AGI-3 is considered one of the most demanding general intelligence tests for AI systems.

More on this topic

💻 Technology OpenAI's AI models hacked Hugging Face servers to cheat on a benchmark Android Authority · 5d ago 💻 Technology Sam Altman to Brief Trump Administration and Senators on OpenAI's Upcoming AI Models CNBC · 3h ago 💻 Technology New York Luddite gnomes staged a mock trial of OpenAI and Sam Altman 404 Media · 7h ago

More on this topic

💻 Technology OpenAI's AI models hacked Hugging Face servers to cheat on a benchmark Android Authority · 5d ago 💻 Technology Sam Altman to Brief Trump Administration and Senators on OpenAI's Upcoming AI Models CNBC · 3h ago 💻 Technology New York Luddite gnomes staged a mock trial of OpenAI and Sam Altman 404 Media · 7h ago