STRONG SIGNAL: OpenAI let a fourth GPT-6 model slip, code-named Astra Minor
On the same day it launched GPT-6 Sol and GPT-6 Luna, OpenAI appears to have accidentally revealed a fourth member of the GPT-6 family. The name first surfaced on September 23, when observer kimmonismus spotted a reference to astra-minor alongside gpt-6-sol and gpt-6-luna in material being prepared for that day's launch. Hours later, the account AiBattle posted a screenshot of OpenAI's own Help Center listing Astra Minor next to GPT-5.5, GPT-5.6 Sol, GPT-6 Sol, GPT-6 Luna and Astra as one of the mainline models under the same Daybreak Blue access framework. OpenAI pulled the reference from public documentation shortly after, which usually signals something shipped early, not that the name was a joke.
Two independent technical sources, not a retweet bubble
What separates this from an ordinary rumor is the convergence of artifacts from different places. On one side, the Help Center screenshot shows Astra Minor classified as standard mainline usage, carrying the same safeguards as regular Astra and no access to cyber specialized models. On the other, the account BridgeMind reported seeing gpt-6-sol, gpt-6-luna and gpt-6-astra-minor show up together overnight in Azure configuration, in the same commit where OpenAI reportedly demoted the older GPT-5.6 Sol to a workhorse category, the kind of relabeling that usually only happens once the replacement is production ready. Support documentation and cloud configuration are two different systems, maintained by different teams inside OpenAI: when both leak the same name on the same day, the correct read is strong signal, not stray gossip.
Where it would sit inside the GPT-6 family
The GPT-6 family today comprises Astra, the flagship launched on September 3 with cyber capability rated critical by OpenAI itself, and the Sol and Luna pair, launched September 22 as cheaper, faster models reusing Astra's training. Astra Minor, by its name and its position in the Help Center list, would not be another cost tier like Sol and Luna: it showed up listed next to Astra itself, as a separate mainline model, under the same Daybreak Blue safety umbrella. The reading most consistent with the leaked material is that OpenAI is testing something that occupies the space between full Astra's frontier intelligence and Sol's lower cost, keeping the refusal standard and safeguards of the flagship product rather than the economy tiers.
The link to the always on assistant that also leaked this week
The timing is not an isolated coincidence. Two days earlier, a separate batch of leaks pointed to an always on assistant inside ChatGPT Pro, internally named o, tied to an infrastructure codename called Aeon and designed for long running tasks that keep executing after the user closes the app. Reporters following that thread speculate Aeon could be precisely an Astra variant optimized for that kind of task, which would make Astra Minor a natural candidate as the engine behind the always on agent product, rather than just another option in ChatGPT's model picker. Neither thread has been officially confirmed, but both point in the same direction: a mid tier model inside the Astra family, built to run longer processes at a lower cost than the flagship.
What is still missing to become fact
There is no model card, no pricing page, no confirmed context window or benchmark, and OpenAI has already shown it knows how to erase a trail fast when something ships too early. It is entirely possible Astra Minor never becomes a public product, stays just an internal route name, or ends up absorbed into whatever the company decides to call o. The recent precedent helps calibrate expectations: the Sol and Luna pair itself first leaked through Codex before becoming an official launch, and the pattern repeated here with one important difference, this time the material came out of two distinct production systems, not a single app screen.
MaxAssistant's read
OpenAI's DevDay is set for September 29, one day after this publication, and is the obvious stage for any real announcement. If Astra Minor shows up there with a name, price and model card, the practical effect is to give OpenAI a fourth cost rung inside GPT-6, slotted between Sol and Astra, competing directly in the mid tier that Grok 4.7 and Gemini 3.8 Flash are already fighting over. If it does not show up, the name stays on record as one more internal configuration label that leaked early without becoming a product. Either way, the pattern of leaking through dual technical artifacts, documentation plus cloud infrastructure, is today's most reliable way to spot a model launch ahead of the official announcement, and it is worth watching closely over the next 48 hours.
Sources
AiBattle (X), OpenAI Help Center screenshot listing Astra Minor as a mainline model: https://x.com/AiBattle_/status/2102757195490427211 | Chubby / kimmonismus (X), first public mention of the name astra-minor alongside GPT-6 Sol and Luna: https://x.com/kimmonismus/status/2102378736658350468 | BridgeMind (X), report of gpt-6-astra-minor in Azure configuration the same night of the launch: https://x.com/bridgemindai/status/2102366589337153589