Tibo’s 28-day Codex challenge ends with GPT-6.2 Astra rumors
What Tibo’s 28-day Codex challenge delivered, why fans wanted more, and what GPT-6.2 Astra and Claude Fable 5.5 rumors mean amid Anthropic’s IPO plans.
For 28 days, Codex users had a reason to check Tibo’s feed: a product improvement, a usage reset, or both. As the final days approached, some hoped for something bigger—a new OpenAI model to answer Claude Opus 5.5.
The challenge ended on November 1 with a banked reset, not a frontier-model announcement. Around the same time, screenshots of an unfamiliar gpt-6.2-astra label appeared in usage histories. Meanwhile, rumors of Claude Fable 5.5 gave those waiting another possible rival to watch.
By November 2, neither name had a confirmed public launch. The challenge had delivered updates; the hoped-for finale was still missing.
What Tibo promised in the 28-day Codex challenge
On October 4, OpenAI’s Thibault “Tibo” Sottiaux promised that for the next 28 days the team would either ship an improvement relevant to most Codex and ChatGPT Work users or give a full usage reset. A reset refreshes a covered allowance; a banked reset can be saved and redeemed later, subject to its terms.
The series mixed practical fixes with resets and model updates. GPT-6 reached Chat, Terra returned, and the team worked on speed, context and reliability. What did not arrive at the finish was the new flagship some followers had imagined.
Codex challenge summary: resets and feature updates
The announcements fall into a few practical groups. Resets offered temporary relief; the feature and allowance changes affected how people could keep using the product.
| Area | Main changes | What users gained |
|---|---|---|
| Usage resets | Repeated allowance refreshes, ending with two banked resets. | More time to work; banked resets could be saved for later. |
| Speed and reliability | Faster Astra/Sol generation, instant steering, Sol ultrafast and fewer capacity errors. | Less waiting and quicker course corrections. |
| Longer, more efficient work | Sol’s default context grew to 500k; better context reuse and longer caching. | Longer sessions with less repeated processing. |
| Models and access | GPT-6 in Chat and GPT-6.1 Terra’s return. | Broader access and another model option, but no new flagship at the finale. |
| Lasting allowance changes | Higher Pro allowances and cancellation of the planned Pro 200 cut. | Changes to ongoing capacity beyond one-off resets. |
| Everyday features | Message editing, composer predictions, mobile account switching, dot texting and CLI updates. | More convenient daily workflows. |
| Developer tools | Decisions API reached general availability; API and auto-review changes. | More options for building and supervising agent workflows. |
Sources: announcement text collected in Nepenthe’s daily log and TiboSay, plus the community discussion.
A daily schedule also creates a daily audience. Independent trackers appeared, users compared updates, and speculation gathered around the final days. Tibo’s tease about saving the last two resets encouraged that anticipation, although his original pledge never promised a flagship launch. The series worked as marketing as well as product communication: it kept people returning, but also gave their expectations time to grow.
Why some OpenAI fans wanted an answer to Opus 5.5
The response was mixed. In one X discussion, Token Gremlin said they appreciated the updates but felt something was unfinished, later naming a model above the Terra tier as something they wanted. Replies ranged from appreciation for the resets and shipped changes to disappointment that no bigger model arrived.
Those replies are individual reactions, not a representative survey. They show why the same month could feel generous to one user and underwhelming to another: a fresh allowance helps someone hitting limits today, while a developer waiting for a stronger coding model is looking for a different kind of improvement.
There was a concrete reason to want more. Anthropic launched Claude Opus 5.5 on September 22. Anthropic says it matches GPT-6 Astra on Terminal-Bench 4.0 for roughly 40% of the cost in its evaluation; that is the company’s reported comparison for a specific benchmark and settings, not proof that one model wins every task.
GPT-6.2 Astra and Claude Fable 5.5: what is confirmed?
On November 1, users shared screenshots showing gpt-6.2-astra alongside gpt-6-astra in usage history, while their model selector still showed Astra. One shared response snippet also carried the name. The screenshots have not been independently authenticated against the accounts, and the plan tier is not established for all of them; there is no verified basis to say the label is exclusive to $500 subscribers.
A label could reflect a test, a routing or reporting change, or something else. It does not establish the model’s capabilities, public availability or release date. As checked on November 2, OpenAI’s model catalogue did not list GPT-6.2 Astra.
The anticipation was not confined to OpenAI. A LiteLLM change briefly added an Azure entry named claude-fable-5-5 in October; a later removal and independent tracking show why a catalog entry alone cannot establish a launch. As of November 2, Anthropic’s model documentation lists Fable 5.1, alongside the released Opus 5.5. Fable 5.5 has not been announced there, and no performance comparison with it can be made from a catalog label.
Anthropic’s IPO adds pressure to the model race
There is more at stake than a leaderboard. Anthropic filed for an IPO on October 26, while OpenAI has said it does not plan to go public in 2026. In September, Reuters reported that Anthropic was considering a model release amid competitive pressure from Astra and an expected IPO. That report preceded Opus 5.5; it did not establish a Fable 5.5 release or a launch timed to the filing.
Competitive pressure does not remove the practical obstacles to shipping. Safety, security and commercial interests also overlap in Anthropic’s responses to open-weight rivals. Reuters reported from the IPO filing that Anthropic had paused pre-training of a next-generation model for more than a month to strengthen protections against theft. The disclosure does not identify that model as Fable 5.5 or explain OpenAI’s plans.
For users, the distinction is simpler: does the tool help them do better work, and does their subscription last long enough to finish it? The challenge improved parts of that experience. But for those waiting for a stronger flagship, another reset could only extend the wait. Usage labels and launch rumors now carry the anticipation forward; the next model will have to justify it in actual use.
Reporting checked November 2, 2026. X posts were reviewed through indexed mirrors and reporting; direct X access was unavailable.
Loading content…