Reports indicate openai's GPT-6.2 astra may be in limited testing, visible in codex usage logs

Evidence suggests that OpenAI's next-generation model, GPT-6.2 Astra, might be undergoing limited testing, as some Codex users have observed a distinct model identifier, 'gpt-6.2-astra,' within their usage statistics. This observation comes from reports by TestingCatalog, which noted the new identifier in user analysis pages, with one account showing a 19.8% call volume attributed to the model in October's statistics. TestingCatalog posits this indicates a gray or A/B test where new models are routed in the background without changing the visible interface name. While OpenAI has not officially confirmed the existence or release date of GPT-6.2 Astra, the background context involves the prior release of GPT-6 Astra on September 3, 2026. In related benchmarks, GPT-6 Astra scored 57.9% on Terminal-Bench 4.0, compared to Claude Fable 5.1's 55.8%, though the official leaderboard shows slightly different figures. This development coincides with rumors regarding Anthropic's next-gen models, including potential routing of Claude Opus 5.5 requests to an unannounced Fable 5.5, alongside hidden identifiers like Mythos 5.5 in Anthropic's code. Furthermore, some Codex users have reported increased token consumption rates for GPT-6 Astra, which may be linked to the potential GPT-6.2 Astra testing. Consequently, user focus is shifting to the practical aspects of GPT-6.2 Astra, including its actual operational costs, token efficiency, and subscription quota usage, beyond its reasoning and coding capabilities.