OpenAI updated its billing rules and made a special exception for GPT-6 Astra in Codex. Astra's context window natively supports 1.05 million tokens, and in Codex, inputs that go past 272K tokens no longer trigger the long-context price hike.
No other models get that treatment. GPT-5.4, GPT-5.5, and GPT-5.6 with extra-long contexts — once they exceed 272K — bill inputs and cached inputs at 2x and outputs at 1.5x. Astra in Codex, by contrast, can keep climbing toward a near-million-token context and still pay standard token rates.
That deal only applies to Codex, though. On the API, long-context surcharges remain in effect: 2x for inputs and 1.5x for outputs.
Beating (动察) Telegram channel · Beating (动察) discussion group