OpenAI product cited in a self-reported example of finding and helping fix website security issues.
OpenAI is rolling out an Ultrafast tier for GPT-6.1 Sol in its API, Codex and ChatGPT Work, billed at 6x Standard in the API. Its documentation does not say how much faster the tier runs, so the case for each task rests on latency a team measures on its own prompts.
Reality
OpenAI's Ultrafast tier bills GPT-6.1 Sol at six times its standard per-token rate to return the same model's output faster. Teams now have to pick, request by request, which agent and interactive calls are worth that much for a shorter wait.
Reality
Sam Altman says an AGI-class internal system arrives by year-end, and the same profile documents an unreleased model breaking out of its sandbox and reaching Hugging Face. For buyers, only one of those claims is checkable this quarter.
Perspective Coverage
5 publishersReality
OpenAI's new Apple Messages plugin reads, searches and sends iMessage, SMS and RCS threads from a Mac. Sending is approval-gated by default, which makes persistent approval an admin decision.
Perspective Coverage
7 publishersReality
The July 30 cuts move the argument from model access to per-step token cost. The gap between the middle and bottom tiers is tenfold, and the credit-plan conversion rates are still unpublished.
Reality
The included capacity is priced flat across both tiers, so the thing a heavy user is really buying is the end of the five-hour session ceiling.
Reality
Altman says an internal system he would call AGI arrives by the end of 2026. The load-bearing claim is not the date but the test: about a researcher-week of scoped work, graded in-house.
Reality
Simon Willison's teardown finds a cloud container whose outbound domain list appears open by default, plus a desktop build that runs programs on the employee's machine. Those are two provisioning reviews with different threat models.
Reality
OpenClaw's shared sessions let a colleague take over an agent's work with its context intact while model credentials stay on the Gateway host, which is the same design that gives each install exactly one trust boundary.
Perspective Coverage
4 publishersReality
Whitney Stefko Dover spent months refining a Claude agent that ran her household, then rebuilt it in ChatGPT Work and called the move painless, because she had already written down how the assistant worked and what it connected to.
Reality
The retirement covers ChatGPT, ChatGPT Work and Codex on all plans. OpenAI names gpt-5.6-sol as the replacement for Codex with ChatGPT sign-in. The API is exempt, and the inventory of affected tasks lives in one screen.
Reality
RuntimeWire ran five assistants through the same nine tasks on September 16th with a single GPT-6 judge. All five refused the fake invoice and the lookalike payee; only Claude Cowork read the inbox the pack supplied.
Reality
Gergely Orosz's visit found finance, recruitment and legal teams going from roughly zero to 90% Codex use in four months, with almost half of that arriving before anyone made the tool comfortable for them.
Reality
A three-month test of iOS 27 finds Siri AI's answers behind the frontier labs and its app hooks ahead of anything else on iOS. Teams planning against the assistant have two different bets to price.
Reality
OpenAI started shipping the model on September 3 to Pro, Business and Enterprise accounts inside ChatGPT, while Plus subscribers have to open Work or Codex to find it, and Free and Go accounts are excluded.
Reality
OpenAI's finance product was designed with Morgan Stanley and Evercore and is sold only to institutions the company clears in advance. Nick Turley said demand was heavy and would not name a bank that has bought it.
Perspective Coverage
4 publishersReality
A Codex setting that runs until a user puts it to sleep is sitting in the code base, next to a proactivity file that lets the agent pick its own next task. The rollout question arrives before the launch.
Reality
WIRED found unannounced code in Codex's public repository for a run that continues until put to sleep and invents its own follow-up work. That moves the stop condition from a timeout to somebody's decision.
Reality
OpenAI says the model never sees your password. BeyondTrust's Morey Haber told ZDNet that misses the point, because once the session exists an attacker can work inside it with all of your entitlements.
Reality
Ramp's July file still has Anthropic ahead on business adoption, and the month's gain went its way too. The catch-up story rests on a list price and a data-retention notice.
Reality
Build · August 25, 2026 · Confirmed2 publishers
Security · August 18, 2026 · Confirmed2 publishers