GPT-6 shows significant performance jumps in AI benchmarks

OpenAI’s latest model, GPT-6, has demonstrated significant performance leaps in specific benchmarks, most notably the ARC-AGI-3 task. In this test, which evaluates how agents learn to solve unfamiliar interactive tasks, GPT-6 achieved a score of 99.9%, compared to 30.2% for Anthropic’s Opus 5 and 7.8% for GPT-5.6.

While industry leaders like Nvidia CEO Jensen Huang and OpenAI President Greg Brockman suggest the arrival of the Artificial General Intelligence (AGI) era, some users argue this characterization is an exaggeration. Despite the debate over the definition of AGI, GPT-6 shows practical improvements in coding efficiency, speed, and agentic capabilities, such as identifying errors in complex development formats that previous models missed.