OpenAI's Codex team locks in for a 28-day shipping sprint

If you notice developers burning through their AI allowances faster than usual this month, they might just be trying to win a bet against OpenAI.

Thibault Sottiaux, engineering lead for Codex at OpenAI, has set an unusually steep challenge for his team. Over the next 28 days, the crew must ship a clear, user-facing improvement for most Codex and ChatGPT Work users every single day. If they miss even once, OpenAI will trigger a full usage reset for subscribers.

Fast tokens and high-stakes wagers

Day 1 of the sprint is already in the books, meaning anyone holding out for an early freebie will have to wait. The team kicked off the challenge by speeding up GPT-6 Astra and GPT-6.1 Sol by roughly 50%, pushing output from about 30 tokens per second to roughly 50. The boost also applies to third-party setups connecting via ChatGPT accounts, including Devin, OpenCode, Pi, and Amp.

Sottiaux noted the sprint will focus strictly on four priorities: simplifications, greater efficiency that unlocks more usage, breakthrough features, and brand-new models.

The bet took off instantly across the developer community. Within hours, an unofficial tracker popped up to grade each day. Several coders joked that Sottiaux had effectively turned usage limits into a daily casino, encouraging people to exhaust their quotas every morning in hopes that the Codex team misses a deadline.

An awkward calendar and the Opus problem

Resetting usage limits has become a signature move for OpenAI. Sottiaux handed out resets when Codex hit 8 million users, did it twice in one day after a buggy July desktop integration, and joined Sam Altman to issue another on stage at DevDay on September 29.

Yet the timing of this charm offensive feels a bit uneasy. Starting October 30, right near the tail end of the 28-day window, OpenAI plans to slash the Codex allowance on its $200 Pro plan from 20 times the Plus limit down to 10 times, nudging heavy users toward a new $500 monthly tier.

At the same time, Anthropic is breathing down OpenAI's neck. Influential developers like the creator of T3 Code, Theo Browne, reported that Claude Opus 5.5 captured over 50% of all prompt traffic on the platform as of October 3. As one commenter put it, shipping two dozen incremental tweaks won't matter much if OpenAI cannot deliver a model that tops Opus.

Why it matters

For regular developers and ChatGPT Work subscribers, this sprint is an immediate win. Whether OpenAI ships a genuine productivity upgrade or stumbles and awards a full reset, users get either better software or free compute every 24 hours.

From an industry standpoint, we see this as a loud admission that Anthropic is winning over the coder demographic. Turning product development into a transparent, high-stakes sprint is a clever way to dominate developer mindshare and keep eyes on Codex while Claude Opus 5.5 surges.

Still, the gambit carries real risks. Promising daily wins is exhausting, and because the criteria live only on Sottiaux's social posts, arguments over what counts as a clear improvement are bound to happen. More critically, handing out daily freebies won't soften the blow when OpenAI halves its Pro tier limits at the end of the month.