Three Unverified 'GPT' Codenames — Terra, Sol, Astra — Surface on OpenRouter Under OpenAI's Name

Three Codenamed Listings, Zero Confirmation

OpenRouter, the multi-provider model aggregation platform, has published listing pages for three model endpoints — GPT Terra Latest, GPT Sol Latest, and GPT Astra Latest — each tagged with a tilde-prefixed ~openai identifier. The tilde prefix on OpenRouter typically flags a listing as unverified, aliased, or routed through a proxy rather than a directly confirmed first-party integration. OpenAI has made no public announcement, blog post, or changelog entry referencing any of these three names as of this writing.

No pricing per token, context window size, benchmark scores, or training cutoff date is available for any of the three listings. OpenRouter's model pages, which normally surface this metadata once a provider confirms an endpoint, show no such details here — a strong signal that these are placeholder, stealth, or early-access listings rather than fully documented releases.

What Codenames Like This Usually Mean

Stealth or codenamed model listings are not new to OpenRouter or to the broader industry. Providers occasionally route unreleased or A/B-tested checkpoints through aggregators under generic astronomical or elemental codenames — a practice seen previously with unannounced models from multiple labs before their official reveal. The names Terra, Sol, and Astra follow this pattern: short, thematically linked (Earth, Sun, star), and deliberately uninformative about model architecture, size, or capability.

The fact that three separate codenames appeared together, all tagged to OpenAI, suggests either a batch of internal test variants being routed for evaluation, or a naming convention tied to a broader internal project. Neither OpenAI nor OpenRouter has published documentation distinguishing what separates Terra from Sol from Astra — whether they represent different model sizes, different fine-tuning targets, or simply different snapshot dates of the same underlying system.

Why This Matters — And Why Caution Is Warranted

Because none of the three listings carry confirmed specifications, any claims about their capabilities, performance, or intended use should be treated as speculative. Historically, when labs have run stealth models through third-party routers, the practice has preceded an official launch by days to weeks, used for real-world traffic testing before a public benchmark push. It is equally possible these are internal routing artifacts with no near-term public release plans.

Until OpenAI issues an official statement, independent developers and enterprises should not build production dependencies on any of these three endpoints. OpenRouter listings can be deprecated or renamed without notice when a stealth model graduates to a named public release or is retired entirely.

What This Means

This is a signal worth watching, not a confirmed launch. The appearance of three simultaneously tagged, unconfirmed OpenAI endpoints suggests active internal testing — but reporting these as finished products would be premature. The responsible read is: OpenRouter has surfaced three data points; OpenAI has verified none of them. Expect clarity only if and when OpenAI itself publishes a model card, pricing, or benchmark data under one of these names — or a different, more polished name entirely, as has happened before when codenames get replaced at official launch.