OpenAI's 28-day shipping sprint fuels debate over Codex performance vs. usage limits

OpenAI's engineering lead for Codex has launched a self-imposed 28-day challenge to ship a meaningful improvement to its coding models every single day, or else grant subscribers a full usage reset. The announcement has triggered a mix of enthusiasm and skepticism among developers—many of whom are now gaming the odds of an OpenAI miss to pocket a free quota boost.

The first ship: Speed gains on flagship models#

The sprint's opening move delivered faster inference on GPT-6 Astra and GPT-6.1 Sol, with token-generation speed roughly doubling from 30 to 50 tokens per second. Thibault Sottiaux, the Codex engineering lead behind the pledge, framed the speedup as a straightforward metric win, though the exact improvement numbers remain approximate rather than rigorously benchmarked.

The acceleration applies across OpenAI's own platforms and extends to third-party integrations that rely on Sign in with ChatGPT, including services like Devin and Pi. The timing carries particular weight given that Sottiaux had just apologized days earlier for slow rollout performance on GPT-6.1 Sol and issued a compensatory reset—so the sprint's debut improvement directly addresses a recent pain point.

Resets as a gamified stake#

The real tension underlying Sottiaux's bet lies in how OpenAI has conditioned its user base to expect usage resets as rewards. Over recent months, the company has handed out quota refreshes to address service hiccups, celebrate milestones (the 8 million active user mark), and make up for rolling out features that broke existing workflows. Developers have started tracking reset frequency; within hours of the announcement, an unofficial scorecard appeared to monitor whether Sottiaux delivers on all 28 ships.

The calculus is simple: users who burn through their weekly allowance now have an incentive to hope OpenAI stumbles, guaranteeing a fresh bucket of tokens. The gamification has become dark enough that at least one developer joked about treating Codex as a slot machine—a way of highlighting the absurdity of tying service improvements to cached-up credit resets.

The usage-limit frustration underneath#

The sprint announcement arrives at a fraught moment for Codex's pricing structure. OpenAI disclosed last week that it will halve the monthly quota on its $200 Pro plan starting October 30 (roughly mid-sprint), cutting it from 20× the Plus tier's allowance to 10×. The company is also introducing a new $500 tier with 25× the Plus allowance and a faster GPU option.