OpenAI's GPT-6.2 Astra Leaks in Codex Usage Stats Before Official Reveal

OpenAI has not announced GPT-6.2 Astra, but it has already started showing up in front of users. Multiple Codex users report that the model appears in their usage statistics, even though the main interface still labels everything as GPT-6. Astra. The community has been waiting on either a GPT-6.1 Astra or a GPT-6.2 Bel; what actually surfaced is GPT-6.2 Astra.

What the Screenshots Show

The visible evidence is narrow but consistent. On the Codex usage-stats page, the model selector still reads "GPT-6 Astra," but the detailed per-request logs carry the model identifier gpt-6.2. Several users have posted the same screenshot, which suggests a broad gray rollout rather than a single misconfigured account.

The official OpenAI model directory does not list GPT-6.2 Astra yet. The community theory is that GPT-6.1 Astra was pulled over safety issues, and that 6.2 is the fixed version OpenAI decided to ship instead of the cancelled 6.1.

Some users say GPT-6.2 Astra is reachable in the Codex CLI. One circulating screenshot is ambiguous: the system first returned an unsupported-model error, then completed a /review task, while the status bar still reported gpt-6.2-astra. On current evidence, that is not enough to claim the CLI path is working end to end.

What the Rumored Claims Say

The leak thread carries a long list of reported improvements, none of them confirmed by OpenAI:

• A significant jump over GPT-6 Astra across most domains.
• Claude Opus 5.5 reportedly beaten on most benchmarks.
• Pricing materially lower than any competing model.
• Faster inference and lower token consumption per task.
• Large gains in visual and design aesthetic.
• Stronger technical reasoning and a step-change in coding performance.

Treat these as launch-promise territory until OpenAI publishes numbers of its own. The pattern from prior model cycles is that preview leaks and launch keynotes tend to tell a stronger story than the first few weeks of production use.

The Opposing Side Is Also Teed Up

The natural competitor for GPT-6.2 Astra is Anthropic's Fable 5.5, which leaked out of the Azure model catalog a few days ago. Anthropic has also been leaving breadcrumbs elsewhere: Mythos 5.5 was found buried in the company's public HTML, and Fable 5.5 itself showed up in Azure listings before an official announcement. With Anthropic's IPO filing in motion, both companies have reason to put their next model on the table as soon as possible.

If GPT-6.2 Astra and Fable 5.5 both land in the next few weeks, the next round of head-to-head comparisons — coding, cost per task, and reasoning — will be the first real test of whether either company's launch narrative holds up under production load.