128 vs 960—this budget cut is brutal. Luckily they pulled it back; otherwise the top tier would’ve turned into a beggar edition—who could stand that?

Crypto news outlet reports that OpenAI has responded to allegations that its GPT-5.6 SOL was being questioned as “dumber.” It said that adjusting the reasoning budget was only to help investigate the experiment. Users found that the internal reasoning budget for GPT-5.6 SOL across each tier had clearly decreased: the highest tier dropped from 960 to 128. This led to suspicions that OpenAI weakened the model to save compute. Tibo Sottiaux, head of OpenAI Codex, said the team had adjusted the reasoning strength to investigate why the subscription quota was being consumed too quickly. This experiment has been withdrawn, and official confirmation said the model has not been weakened long term. In addition, OpenAI also completed reasoning optimizations, allowing each GPT-5.6 SOL subscription tier to get about 10% more usage.