📉 OpenAI Reduces Context Window for Codex Models
In the openai/codex repository (branch release/0.144), model metadata has been updated, resulting in a reduction of the context window from 372k to 272k tokens for several models, including the GPT-5.x series. The changes were recorded in pull request #33972, which was merged on July 18, 2026.
🌍 A sharp change in context parameters within OpenAI's infrastructure repository may indicate optimization of inference costs or a change in the token management architecture in the new GPT-5 models.
👤 Developers using APIs or tools based on Codex/GPT-5 need to revise their long-prompt management strategies, as the available memory volume for a single request has decreased by approximately 27%.
Source 1: https://github.com/openai/codex/pull/33972/files