Unreleased Frontier Models Discovered in Developer Telemetry

Tech enthusiasts and software developers recently noticed references to unannounced artificial intelligence models tucked inside system usage logs and website code. According to a report by TestingCatalog, some OpenAI Codex users spotted code requests assigned to a non-public version named GPT-6.2 Astra. Simultaneously, code references to a new Claude Mythos 5.5 model were discovered on Anthropic's official website.

For everyday users, these quiet appearances signal that major tech providers are actively testing their next generation of tools in real world conditions. Rather than holding massive unveilings before any public access, companies frequently route small portions of active user traffic to newer models to evaluate performance improvements, speed, and safety. This suggests that faster or more capable updates to consumer AI assistants could arrive within days or weeks.

Despite the technical clues left in data dashboards, neither OpenAI nor Anthropic has publicly announced these specific updates. It remains uncertain whether these telemetry sightings represent final release candidates or experimental versions that might still be modified or delayed. This quiet deployment strategy raises an interesting question about transparency, as users may be evaluating software outputs without knowing whether they are interacting with established tools or unannounced experimental systems.