OpenAI Revises Astra's Performance Metrics Post-Launch, Raising Questions on Benchmark Integrity

OpenAI altered several evaluation benchmarks for its GPT-6 Astra model after the initial blog post publication, with some adjustments improving Astra's scores and lowering those of rival Anthropic's models. The post experienced a delayed rollout and was retracted before being republished with revised figures. OpenAI stated the changes were made to ensure the numbers represent the best estimate of model performance.