OpenAI stretches Codex usage with subagent context reuse and longer caching

OpenAI has changed Codex and Work so subscribers get more effective usage out of the same plan.

Codex lead Thibault Sottiaux said on X on Oct 16 that the team now reuses context across subagents, keeps context cached longer and routes requests to raise cache hit rates. He did not say how much usage grows. The post is "Day 12" of a running series of fixes; on Oct 15 he said the team had sharply cut "model at capacity" errors. It drew 3,300 likes. Theo of t3.gg praised the transparency, while one user replied that "you still can drain your 7d in 1 day."