OpenAI published 99.9%; benchmark software returned 62.7%, The Next Web reports

The Next Web reports that OpenAI published a 99.9% benchmark result, while the benchmark’s own software returned 62.7%. It describes the same model and test with different surrounding software setups, and says ARC Prize printed both numbers without claiming artificial general intelligence (AGI). A critic sharing the report accused OpenAI of pretending Astra was AGI and called for the company to be shut down until there are “changes at the top.”