Three unreleased models are circulating as imminent right now, and the evidence behind each is wildly different.

GPT-6 Sol is rumored as a cheaper sibling to the GPT-6 Astra flagship OpenAI shipped this month at $10 and $50 per million tokens. Claude Opus 5.2 is supposedly being routed inside Claude Code already, with a release this week. And Grok 5 has been coming soon since January, with Musk describing it as a step toward AGI.

One of those has a shipped precedent behind it. One traces to a single X account and a timeline you can watch escalate hour by hour. One has missed two windows confirmed by the company's own founder.

Same phrase, three completely different things. Here is how to tell them apart.

The case for GPT-6 Sol does not rest on a leak. It rests on what OpenAI did three months ago.

The GPT-5.6 generation shipped as a family rather than a single model. GPT-5.6 Sol was the flagship at $4 and $20 per million tokens, with Terra and Luna stepping down for mid-tier and high-volume work. It did not arrive all at once either. Sol went through a limited federal-review preview starting June 26, then the full set went public on July 9.

GPT-6 Astra shipping first at premium pricing with cheaper siblings following would be that identical playbook, one generation later. That is pattern recognition rather than speculation.

The problem is the name, and it is a real one. OpenAI has sold GPT-5.6 Sol since July 9 and it is still on the price sheet alongside GPT-6 Astra. Shipping a GPT-6 Sol now would put two models called Sol on the same page, at different generations and very different prices.

So the tier is close to certain. The name is the part to doubt.

This one is worth reading closely, because someone documented the whole thing with timestamps.

At 07:57 UTC on September 14, one X account listed Opus 5.2 in a launch-week wish list, alongside Gemini 4, Grok 4.7, and the GPT-6 family. A wish list.

At 12:15 UTC the same account wrote that Opus 5 was getting routed to Opus 5.2 inside Claude Code, because one prompt on the highest effort setting came back cleaner there than in Devin.

At 14:00 UTC that became Opus 5 is confirmed routing Opus 5.2, with a test prompt attached for others to run.

At 14:38 UTC it became mostly coming this Thursday or Friday.

Six hours and forty-one minutes from wish to confirmed. By mid-afternoon a dozen accounts had repeated it and Polymarket had moved. None of them added a document.

As of this week Anthropic has published nothing. No announcement, no model card, no API identifier, no pricing, no documentation entry. Its newest Opus remains Opus 5, released July 24 at $5 and $25 per million tokens with a one-million-token context. Its newest model overall is Claude Fable 5.1, which shipped September 1.

Some outlets have since run this as a launch, reporting that Opus 5.2 is live in limited beta with faster responses and improved code output. Those reports trace back to the same thread. Polymarket now has the next Opus at 80 percent by September 30, which may well be right, because Anthropic does ship frequently. But a prediction market pricing a probability is not the same as evidence a specific model exists.

Grok 5 has been coming soon for three quarters, and the substance behind it has not changed.

Musk originally confirmed Q1 2026. That passed. xAI's account then pointed at Q2. That passed too. Expectations moved to a May or June beta with Q3 API access. It is September, and xAI's own documentation still lists Grok 4.6 as the newest model you can actually use. The official line has not moved in months: Grok 5 is in training.

What fills the gap is positioning rather than specification. Musk has described Grok 5 as a step toward AGI rather than another language model. That framing is the most consistent thing attached to it, and it is not a claim anyone can check.

The rumored specs are ambitious and unconfirmed: roughly 6 trillion parameters in a mixture-of-experts design, double Grok 4's estimated 3 trillion, training on the one-gigawatt Colossus 2 cluster in Memphis, a 1.5-million-token context, native multimodality across text, images, audio, and real-time video, and a multi-agent system scaling from 4 agents to 16 or more. Almost every number there sits in the rumored column.

Meanwhile xAI has been shipping constantly, just not this. Grok Voice, Grok Imagine, Grok Bot, and point releases through Grok 4.6, which landed July 8 at $2 and $6 with a strong Artificial Analysis Intelligence Index score. Each of those keeps getting folded into Grok 5 is here headlines that are simply wrong.

Treat these differently when you plan.

A cheaper GPT-6 tier is a reasonable assumption for the next few months, because the pattern is established and the economics demand it. Keep model IDs configurable and you can switch the day it lands.

Opus 5.2 might well arrive this month. Anthropic ships often, and Fable 5.1 landed September 1. But nothing about the current rumor constitutes evidence, and if you are making a decision this week, Opus 5 at $5 and $25 is what exists.

And if you are waiting on Grok 5, stop. Grok 4.6 is live at $2 and $6 with a strong independent score. Building a plan around a model that has missed two company-confirmed windows is how quarters get wasted.

The broader point is that coming soon carries almost no information by itself. What carries information is whether the company has done this before, whether you can trace the claim to a document, and whether the date has already slipped once. All three are checkable in about a minute, which is less time than most people spend arguing about the rumor.

The stories moving fast beyond today's headlines:

Three models, three very different reasons to believe them.

GPT-6 Sol is the safe bet. OpenAI just ran this exact play with GPT-5.6, flagship first at $4 and $20, then Terra and Luna underneath. Astra shipped at $10 and $50, so cheaper siblings are close to inevitable. The weird part is the name, since GPT-5.6 Sol is still on sale and two Sols on one price sheet would be a mess.

Opus 5.2 is one guy on X. At 07:57 UTC it was on his wish list. By 14:00 it was confirmed. By 14:38 it was shipping Thursday. Six hours and forty-one minutes, a dozen accounts piling on, Polymarket moving, and nobody produced a single document. Anthropic's newest Opus is still Opus 5 from July.

Grok 5 has been three months away since January. Musk called Q1, then Q2, and xAI's docs still say Grok 4.6. What we get instead of a date is Musk calling it a step toward AGI, which nobody can check. Every spec floating around is rumored.

Stay building. 🤖