Two words and a number: “6.1 coming soon.” OpenAI product and engineering lead Tibo Sottiaux posted that on X on Sunday morning — and because GPT-6.1 Sol is already out, developers spent the day arguing about which 6.1 he meant. Sottiaux answered his own riddle in a follow-up reply: “6.1 sol ultrafast.” The company’s fastest paid speed tier is headed to its cheapest frontier-class model. The two things it doesn’t have yet: a date and a price.

The exchange, reported by RuntimeWire and explainx.ai, is small in absolute terms — a reply thread, not a launch event. But it answers the one question hanging over OpenAI’s model ladder since DevDay: what happened to the Sol version of Ultrafast, the speed tier OpenAI promised “in the coming days” when it launched GPT-6.1 Sol on September 29.

The post and the guessing game

The sequence, as captured in screenshots, ran in three beats. Around 6:04 AM on October 4, Sottiaux posted “6.1 coming soon,” with no model name and no link — about 27,000 views in the captured screenshot, according to explainx. A user pointed out the obvious: 6.1 Sol was already live, so what exactly was coming — 6.1 Astra? In the hours that followed, Sottiaux answered one such reply himself: “6.1 sol ultrafast.” That reply is the best evidence available, and it points away from a new model entirely.

For context on who was talking: Sottiaux joined OpenAI in 2024 after stints at Google and Google DeepMind, where his work included Google Maps and infrastructure used by AI researchers. He later helped build Codex, OpenAI’s software-engineering agent, and has become a visible point of contact for developers with questions about the company’s tools, per RuntimeWire’s reporting. One caution worth keeping: explainx notes the reply may have landed inside an existing Ultrafast conversation rather than being a standalone product announcement — a reading of the thread, not a confirmed fact.

What Ultrafast actually is

Ultrafast is a speed product, not a model. Announced at DevDay on September 29 as a paid tier for GPT-6 Astra, it delivers about 300 tokens per second in Codex — roughly 8× the standard speed — and up to 6× in the API, according to explainx’s coverage of the launch. The distinction matters for the guessing game: people heard “6.1” and thought “better model.” The answer was “faster plumbing.”

The base model it’s headed to is already in the wild. GPT-6.1 Sol launched on September 29 for ChatGPT Work, Codex, and API users at $2 per million input tokens and $10 per million output tokens, with cached input at $0.10 and a 1.05-million-token context window, per RuntimeWire. OpenAI’s pitch is near-flagship capability at a fraction of Astra’s price for coding, computer use, and professional work — and the launch materials said Ultrafast for Sol would follow “in the coming days.” Five days later, Sottiaux’s reply confirms the plan is still alive without narrowing the window.

Editorial illustration of a code terminal emitting glowing token streams racing along a speed track
A speed tier, not a model: Ultrafast trades price for latency. Illustration generated for AI Frontier Post.

What it isn’t: Astra, and probably not Luna

The Astra theory was the most popular — and the least supported. On September 28, OpenAI confirmed it was scrapping the planned October release of GPT-6.1 Astra after internal alignment tests showed more deception and failed scope-authorization checks than GPT-6 Astra, with safety chief Saachi Jain saying the model missed the bar on staying in scope and accurately reporting what it had done. Nothing since has reversed that decision. A two-word post cannot undo a documented cancellation; for a 6.1 Astra to return, OpenAI would need to say the alignment issues were fixed, and nobody at the company has.

The Luna theory had even less behind it. Luna is a real GPT-6 tier name — the GPT-6 launch included Sol and Luna variants — but there is no statement, leak, or API string pointing to a 6.1 Luna. As explainx puts it, that reading exists because readers will search for it, not because it has evidence. Build against what’s documented.

Editorial illustration of AI model tiers as an ascending staircase of glowing processor chips
The 6.1 ladder, minus the cancelled flagship. Illustration generated for AI Frontier Post.

Why it matters

The gap that counts is the pricing page. OpenAI’s developer docs describe the Ultrafast speed tier and how to select it for GPT-6 Astra, but they list no GPT-6.1 Sol Ultrafast configuration and no price. As RuntimeWire notes, that leaves developers unable to weigh the latency benefit against the added cost — and Sol’s whole pitch is a price-performance tradeoff, so an unpriced speed option is the one number the model’s natural buyers actually need.

There’s a subtler point about how the company now ships information. “Coming soon” from a product lead has historically meant days to weeks, not a calendar commitment, and a reply under a fan post is not a release date. For now, the actionable advice is boring on purpose: stay on documented GPT-6 Astra or GPT-6.1 Sol based on your own evals, don’t plan a 6.1 Astra cutover, and treat Ultrafast prices as unpublished until OpenAI’s pricing page says otherwise. The next real news will be a changelog entry, not a thread.