A new neural network under the codename Polaris Alpha has been released for free testing on the OpenRouter platform. This “stealth model” does not disclose its developer, but within the IT community, there is active discussion that it may be an early test version of GPT-5.1 from OpenAI. The main argument supporting this theory is its knowledge cutoff, which matches only the GPT-5 family.
Polaris Alpha hides information about its creator, but it reveals the date up to which it has knowledge: October 2024. Among all models currently available on the market, this data cutoff aligns only with the GPT-5 series.
In addition, test queries sent to Polaris Alpha show a style of code generation and response formatting very similar to what was seen in earlier leaks of the GPT-5 Thinking model.
The appearance of Polaris Alpha on a testing platform coincides with rumors of an upcoming release of GPT-5.1. References to this new OpenAI version have already been discovered several times in the code of the Codex service.
One of those references even mentioned a possible release date: November 24, 2025. While this date is unusual for a major OpenAI launch, the frequency of recent version updates (such as GPT-4 and GPT-4o every 2–4 months) makes an imminent release quite plausible. The public testing of Polaris Alpha only strengthens these expectations.
ORIENT
