The short version
- Codex Usage Logs Reveal Unannounced 'gpt-6.2-astra' Handling 20% of Tasks (Spotted Identifier: gpt-6.2-astra (alongside gpt-6.2-betelgeuse); Discovery Location: OpenAI Codex account Usage logs).
- Why OpenAI Runs 20% Live Repository Canaries Before Launching Coding Models (Canary Traffic Share: ~15% to 20% of live coding tasks on enrolled accounts).
- How to Check Your Own Codex Usage Breakdown for Astra Routing (Current Status: Silent production canary (unannounced)).
At a glance
| Spotted Identifier | gpt-6.2-astra (alongside gpt-6.2-betelgeuse) |
|---|---|
| Discovery Location | OpenAI Codex account Usage logs |
| Canary Traffic Share | ~15% to 20% of live coding tasks on enrolled accounts |
| Current Status | Silent production canary (unannounced) |
| Category | AI & Models |
| Creator / publisher | Token Gremlin |
| Official link | openai.com ↗ |
| Heat | 🔥 155.5K Heat |
| Announced | |
| Data updated |
Codex Usage Logs Reveal Unannounced 'gpt-6.2-astra' Handling 20% of Tasks
Developers inspecting their OpenAI Codex token usage logs discovered an unannounced model identifier, gpt-6.2-astra, quietly handling roughly 20% of their coding tasks alongside separate sightings of gpt-6.2-betelgeuse.
Why OpenAI Runs 20% Live Repository Canaries Before Launching Coding Models
Routing 15% to 20% of live repository editing and patch-generation tasks to a candidate checkpoint is a classic canary evaluation method used to measure patch pass rates, tool-call stability, and latency before a broader launch.
How to Check Your Own Codex Usage Breakdown for Astra Routing
While OpenAI has not published a standalone model card for gpt-6.2-astra yet, Codex subscribers can check their own enrollment by opening the Usage tab and grouping the last 7 days of token consumption by model name.
FAQ
What specs and pricing details surfaced in the OpenAI gpt-6.2-astra leak?
Spotted Identifier: gpt-6.2-astra (alongside gpt-6.2-betelgeuse); Discovery Location: OpenAI Codex account Usage logs; Canary Traffic Share: ~15% to 20% of live coding tasks on enrolled accounts. Routing 15% to 20% of live repository editing and patch-generation tasks to a candidate checkpoint is a classic canary evaluation method used to measure patch pass rates, tool-call stability, and latency before a broader launch.
How should developers treat these unconfirmed details before an official release?
Current Status: Silent production canary (unannounced). While OpenAI has not published a standalone model card for gpt-6.2-astra yet, Codex subscribers can check their own enrollment by opening the Usage tab and grouping the last 7 days of token consumption by model name.
