AI News

GPT-6.2 Astra Tracker and the Codex Screenshot Evidence

Status as of November 1, 2026: Screenshots posted on X show the label gpt-6.2-astra in usage charts and a JSON-like fragment. OpenAI has not confirmed a GPT-6.2 Astra release in the official sources we checked. The images do not establish what model ran, whether OpenAI is testing one, or when anything might launch.

The label has drawn attention because it appears alongside the documented gpt-6-astra name. If it represents a new model, that could matter to people who use Astra for coding and other demanding work. For now, the useful facts are narrower: people posted images containing the identifier, and several of the posts circulating those images share the same source material.

What the usage screenshots show

On November 1, Mashiro posted three screenshots described as having been collected from three users. Kingy inspected all three images, but the crops do not identify the underlying accounts, so we could not verify that they came from three distinct users.

The first image shows a usage analysis page covering October 26 through November 1. Its legend lists gpt-6-astra at 80% and gpt-6.2-astra at 19.8%. The image does not establish the denominator behind those shares. It would be wrong to read 19.8% as the percentage of users in a rollout or as a verified share of requests or tokens.

Mashiro usage chart showing gpt-6.2-astra at 19.8%

View Mashiro’s first usage screenshot. One of three screenshots collected and posted by @Mashiro_0619. The account data and backend model identity have not been independently authenticated.

The other two images also display the 6.2 label, but their visible date context differs. The crops do not provide enough information to combine their percentages into a meaningful average or estimate the size of any test.

Token Gremlin’s post describes a mismatch between the model picker and the usage label and suggests an A/B test. None of the images we inspected shows the picker, so that mismatch remains the author’s reported observation. In a subsequent reply, Token said, “I used his screenshot,” referring to Mashiro’s image. The two posts therefore cannot be counted as independent screenshot evidence. In a later reply, Token said Mashiro had also reposted the image and that he could not find its original source. We have not independently established that earlier origin.

The JSON image and the reuse chain

A separate post by Salio contains a short JSON-like image. It shows an outer type value of response.completed and a nested response.model value of gpt-6.2-astra.

Salio JSON fragment showing the gpt-6.2-astra label

View Salio’s JSON image. The fragment was reused by other accounts; the crop does not include a full request or response trace and does not independently prove model access.

The image has no visible endpoint, request, headers, response ID, timestamp or reproducible method. It does not tell us whether the fragment was captured from an actual response or whether the identifier is publicly callable. Salio interpreted the image as evidence of a test; that interpretation remains unverified.

Two other posts supplied for this review share the same displayed JSON fragment. kimmonismus’s post attributes the image to Salio on X. Mark Kretschmann’s post shows it as coming from Token Gremlin, whose later post also shared that image. Predictions of an imminent release in those posts are predictions, not launch confirmation.

Evidence reviewed What it supports What remains unverified
Mashiro’s three usage images The posted charts contain the gpt-6.2-astra label. Account identity, the metering denominator and the model behind the label.
Token Gremlin’s usage post The author reports a picker mismatch and shares a crop. The picker claim and proposed A/B test; the author says the image was reused.
Salio’s JSON image, shared by kimmonismus and Mark The posted fragment contains the identifier in a model field. Its origin, full trace, endpoint and whether a real response produced it.

A test, a reporting-label error, a serving alias or a stale interface are possible explanations. The posted material does not distinguish among them. It also does not establish that the label cannot be a glitch.

What OpenAI has documented

OpenAI’s GPT-6 Astra model page documents the identifier gpt-6-astra. In the public API model catalog, API changelog, and Work and Codex models guide reviewed on November 1, we found no GPT-6.2 Astra listing or release announcement.

That is a check of those public sources, not a test of every account or a denial of a private experiment. We did not probe the API for gpt-6.2-astra or run a GPT-6.2-specific model test.

The reviewed sources establish no verified GPT-6.2 Astra release date, access terms, API availability, price, specifications or benchmark results. GPT-6 Astra’s documented details should not be copied into those fields. For readers considering a workflow change, there is no confirmed new model or price here to plan around yet.

Kingy has separate coverage of the Bel and Betelgeuse rumor. None of the sources reviewed for this tracker establishes that Bel or Betelgeuse is the same model as gpt-6.2-astra.

What would change the status

A public OpenAI announcement or model documentation naming GPT-6.2 Astra would establish an official product claim and could clarify access, pricing and timing. A verifiable account-level record with the full context of the usage metric, or a reproducible response trace, could clarify what the current screenshots represent. Neither alone would establish model quality: that needs a clearly identified model and documented testing.