OpenAI Rolls Codex Context Back to 272K — Your Quota Was Paying for It
OpenAI cut the GPT-5.6 Sol context limit in Codex from 372K back down to 272K. The official explanation, from Codex lead Tibo on X: the 372K rollout was one of the main sources of unexpectedly high usage burn. Bigger context meant every request metered more tokens, and subscribers watched quota evaporate without knowing why. The rollback plus new inference optimizations should return roughly 10% more usage; OpenAI says it wants to bring 372K back later.
The community is not fully soothed. A GitHub issue on openai/codex measures the effective window at 258K against an advertised 1.05M, and codex-resets.com — a community site that exists purely to track when usage limits reset — hit the Hacker News front page the same weekend at 287 points. The context rollback thread itself sat at 273. When your users build a countdown site for your rate limits, the pricing model is the product experience.
One practical warning buried in the thread: OpenAI notes that requests beyond 272K are over-charged, and many third-party harnesses still have GPT-5.6's context window configured at 372K. If you run Sol through your own harness, check that setting today — it is silently draining quota. The bigger story: context windows have become a billing lever, and this is the first time a lab has publicly rolled one back to protect users' quota rather than its own capacity.
Statement: https://x.com/thsottiaux/status/2076495156757577895
Tracker: https://codex-resets.com
← Back to all articles
The community is not fully soothed. A GitHub issue on openai/codex measures the effective window at 258K against an advertised 1.05M, and codex-resets.com — a community site that exists purely to track when usage limits reset — hit the Hacker News front page the same weekend at 287 points. The context rollback thread itself sat at 273. When your users build a countdown site for your rate limits, the pricing model is the product experience.
One practical warning buried in the thread: OpenAI notes that requests beyond 272K are over-charged, and many third-party harnesses still have GPT-5.6's context window configured at 372K. If you run Sol through your own harness, check that setting today — it is silently draining quota. The bigger story: context windows have become a billing lever, and this is the first time a lab has publicly rolled one back to protect users' quota rather than its own capacity.
Statement: https://x.com/thsottiaux/status/2076495156757577895
Tracker: https://codex-resets.com
Comments