Codex GPT-6 Astra Rollout Completes With Bonus Usage Resets

CodexView original changelog

Two days after GPT-6 Astra's limited-organization debut, OpenAI finished rolling the model out to every Plus, Pro, and Business plan and made it the bundled default model in Codex CLI (version 0.153.4), so users no longer have to select it manually. The staggered rollout frustrated paying subscribers who expected day-one access, prompting a public apology from Sam Altman and a compensation program of "banked resets", one bonus usage reset for every day a user lacked Astra access. Codex lead Thibault Sottiaux confirmed the rollout finished ahead of the original weekend target and issued the final banked reset the same day.

Key Takeaways

  • GPT-6-Astra became Codex's default model on September 4 via CLI 0.153.4, so new sessions without a saved model preference now use Astra automatically instead of GPT-5.6 Sol.
  • A staggered, multi-day rollout left many Plus, Pro, and Business subscribers without promised access even after the public launch, sparking visible frustration.
  • Sam Altman personally apologized for the "messy rollout," explicitly comparing it to earlier rollout missteps during the GPT-5 launch.
  • OpenAI compensated affected users with "banked resets," granting one bonus usage reset for every day a paying subscriber lacked Astra access, with the first reset landing within hours of the announcement.
  • Codex lead Thibault Sottiaux confirmed on September 5 that the rollout finished ahead of the original weekend target, closing the program with a final full reset for all Plus, Pro, and Business users.
  • Accounts not yet entitled to Astra are advised by OpenAI to explicitly pin GPT-5.6 Sol as a stable fallback until the model reaches their plan.

Astra Becomes the Default Model in Codex

On September 4, 2026, OpenAI shipped Codex CLI 0.153.4, a release whose most consequential line was a single configuration change: GPT-6-Astra became the bundled default model whenever no model is explicitly configured. Until this point, Codex users who wanted Astra had to select it manually from the model picker, which itself had briefly lost the ability to display Astra at all after an earlier 0.153.1 patch (a bug the same release fixed). With the default flipped, any developer starting a fresh Codex session without a saved model preference now lands on Astra automatically rather than on GPT-5.6 Sol, the prior default.

OpenAI's guidance for accounts not yet entitled to Astra is to explicitly pin GPT-5.6 Sol as a stable fallback until the model reaches their plan.

A Rocky Rollout and an Apology

Astra's broader availability did not arrive smoothly. OpenAI had promised access to ChatGPT Plus, Pro, Business, and Enterprise plans "in the coming days" after the September 3 announcement, but many paying subscribers, Pro users in particular, found themselves without access even after the public launch. The mismatch between OpenAI's messaging and what subscribers actually saw drew visible frustration, and CEO Sam Altman posted a public apology, acknowledging the company had made the rollout "messy" and referencing OpenAI's own history of bumpy launches, including GPT-5's. Altman said he hoped access would broaden over the following weekend but stopped short of a firm commitment.

Banked Resets as Compensation

To offset the wait, Codex and ChatGPT lead Thibault Sottiaux announced that OpenAI would grant one banked usage reset for every day a paying subscriber lacked Astra access. A banked reset zeroes a user's usage counter immediately rather than waiting for the normal automatic refill, effectively handing out extra capacity on the models already available to that account. The first reset landed within hours of the announcement, with additional resets following on subsequent days of the delay.

On September 5, Sottiaux announced the rollout had finished ahead of schedule, thanked users for their patience, and confirmed a final full banked reset would land that day for every Plus, Pro, and Business user, closing out the compensation program alongside completed access to Astra.

Why It Matters

Beyond the immediate drama, the episode marks Astra's transition from a narrowly gated preview to Codex's default, everyday model, a materially bigger shift for typical users than the original limited-audience launch. It also illustrates a recurring pattern for major OpenAI model launches: ambitious rollout promises followed by staged delays, public apologies, and goodwill compensation, a dynamic that shaped user sentiment around Astra as much as the model's underlying capabilities did.