Unconfirmed: GPT-6.2 Astra Model ID Appears in Some Codex Usage Data
TestingCatalog reports an unannounced model identifier in Codex usage analytics; OpenAI has not confirmed any GPT-6.2 Astra model.
Published
Last updated / Fact-checked on
A model ID labeled "gpt-6.2-astra" has started appearing in some OpenAI Codex users' usage analytics, TestingCatalog reported on November 2, while the Codex model picker still shows GPT-6 Astra. OpenAI has not announced or confirmed any GPT-6.2 Astra model, so this remains an unconfirmed report.
Photo: Lalmch / Pixabay
The report at a glance
| Model ID seen | gpt-6.2-astra |
| Where | Some users' Codex usage analytics |
| When | Only on the Oct. 31 bar in breakdowns TestingCatalog saw |
| Share of usage | 1.6% to 37.7%, varying by account |
| Model picker | Still says GPT-6 Astra |
| OpenAI status | Not announced or confirmed |
| Possible explanation | A/B test or staged rollout (TestingCatalog's view) |
| Source | TestingCatalog |
Sightings limited to one day
TestingCatalog reported that the “gpt-6.2-astra” entry appeared in screenshots shared by Codex users from the usage page that tracks plan consumption across Work, Codex and other agent tasks. Several users noticed it over the weekend, the site said. In every breakdown the site examined, the model ID showed up exclusively on the October 31 bar. The share of usage attributed to it varied substantially by account: TestingCatalog cited 1.6 percent of one user’s weekly total, 19.8 percent of another user’s total, and 37.7 percent of a single day for a third user.
TestingCatalog interpreted the pattern as evidence that some requests may be routed to a different model without the user interface reflecting the change, describing the situation as either an A/B test or the possible start of a staged rollout. That assessment is the publication’s own interpretation and has not been confirmed by OpenAI.
Raw response data and official silence
TestingCatalog also highlighted a post from X user Token Gremlin that included a raw response event containing the model field “gpt-6.2-astra.” The outlet characterized the snippet as strong evidence that the identifier is live on OpenAI’s servers, while emphasizing that “it isn’t an announcement.”
According to TestingCatalog, OpenAI has not publicly mentioned GPT-6.2 Astra. The company’s model documentation continues to list GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna. Because the Codex picker does not indicate which specific model generated any given response, TestingCatalog stated that it is difficult to confirm whether any particular output originated from a 6.2 model.
Some testers who believe they received GPT-6.2 Astra output have posted results on X, including a 3D pelican-on-a-bicycle scene and a game built from a single prompt, TestingCatalog said. Because the source model cannot be confirmed, those demos are not evidence of what the model can do. TestingCatalog said it leans toward the view that the ID marks a newer, more refined version of Astra, but that is also the outlet's opinion.
Model history and timing
In the same report, TestingCatalog wrote that OpenAI shelved a GPT-6.1 Astra model in late September after internal testing raised concerns about deception and scope. The publication explicitly labeled any suggested connection between that earlier model and the newly observed “gpt-6.2-astra” identifier as “only speculation.” OpenAI has not made that connection.
The outlet also noted the timing. OpenAI released GPT-6.1 Terra a couple of days before the sightings, TestingCatalog said, and Codex lead Thibault Sottiaux's daily shipping series wrapped up with usage resets, which some users are reading as a hint at a launch window. TestingCatalog presented that reading as speculation, not a confirmed plan.
Competitive pressure is part of the backdrop. Google announced Gemini 4 Carbon on October 28, initially for a subset of cyber defenders, and TestingCatalog wrote that Anthropic's Fable 5.5 is already with partners.
Why it matters
These unconfirmed observations matter to Codex users and developers who rely on the tool for coding tasks, because an unannounced change in backend model routing could affect response characteristics without clear indication in the interface. Until OpenAI comments, the practical impact cannot be verified.
What to watch
- An OpenAI statement. Whether OpenAI confirms, renames or denies the model.
- Model docs. Whether GPT-6.2 Astra appears in OpenAI's model documentation.
- Codex picker. Whether the picker starts showing which model answered.
- Usage data. Whether the ID shows up beyond the Oct. 31 sightings.
More coverage is on our Artificial Intelligence hub.
How we reported this
This article relies on TestingCatalog's report and the screenshots and X posts it describes; we did not see the usage data ourselves, and OpenAI has said nothing publicly. Interpretations, including the A/B-test theory, are TestingCatalog's and are attributed.
Published and fact-checked November 3, 2026.
Sources
- TestingCatalog, Dwaybe Cubbins, "GPT-6.2 Astra appears in Codex usage ahead of the launch," Nov. 2, 2026: https://www.testingcatalog.com/gpt-6-2-astra-appears-codex-usage/
- Google, "Introducing Gemini 4 Carbon," Oct. 28, 2026: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.