According to CoinMeta, OpenAI has updated its billing rules. GPT-6 and Astra have made exceptions in Codex, supporting up to 1.05 million token contexts. Even if the input exceeds 272k, it no longer triggers an additional charge for long contexts. Other models such as GPT-5.4, GPT-5.5, and GPT-5.6 will charge double for inputs and cached inputs that exceed 272k, and charge 1.5 times for outputs when using ultra-long contexts. This discount is only applicable to Codex; API will still implement an additional charge for long contexts.