Close Menu
    Special Report

    WELCOME TO THE AGI ERA? OpenAI Just Declared It. The Benchmarks Didn’t. And Elon Says Wait Ten Days.

    superadminBy 12 Mins Read
    “Tired of the same tired news cycle? Escape to the full unfiltered feed at Whatfinger.com — America’s Frontline Truth Feed. 85%+ of readers come direct every day.” –  CLICK HERE for the FULL deal.🛑

    OpenAI President Greg Brockman walked into a press briefing Thursday, talked up the company’s new model, GPT-6 Astra, and then dropped the line Silicon Valley has been teasing for two years:

    “Welcome to the AGI era.”

    That is the quote ZeroHedge ran with. That is the quote ripping across X. And that is the quote every lab in the race now has to live with — because if you say the era has begun, you either own the moment or you own the hangover.

    So let’s do this the Whatfinger way. Not the press-release way. Not the “trust the experts” way. The scoreboard way.

    What OpenAI actually shipped

    Astra is not a chatbot upgrade with a new coat of paint. OpenAI trained it on the company’s largest run yet — more than 100,000 GPUs at the Stargate site in Texas — and it is the first model the company says other models helped supervise at scale. The pitch is blunt: anything you can do on a computer, Astra can do for you. Fast. Legal contracts. PCB layouts in KiCad. 3D scenes in Unity. Tax drafts from a W-2. Prime-number math. Biology and physics evals. And, the part that should make every CISO sit up, previously unknown software vulnerabilities and working exploits against hardened systems without being told where to look.

    That last item is why Astra sat on the runway. It is the first OpenAI model to hit the company’s own “critical” cybersecurity threshold. The public version is supposed to refuse the nasty cyber work. The full kit stays with vetted defenders in the Daybreak program. Regular Plus, Pro, Business, Enterprise, and API users get it “in the coming days.” Price: $10 / $50 per million input/output tokens — 2.5 times the last flagship.

    Sam Altman posted the victory lap himself: best in the world for computer use, professional work, science, coding, and cybersecurity. He cited 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench. Brockman, asked whether this was the AGI model, said, “I think it might be about this model.” Then he welcomed everyone to the era.

    That is the sales pitch. Now the fine print.

    The number they wanted you to share — and the number they hoped you wouldn’t

    ARC-AGI-3 is the sexy one. It is the test built to measure whether a system can walk into a new interactive environment with no instructions, figure out the rules, and act like a human who just got dropped into a strange game. OpenAI’s own headline number is 99.9%. ARC Prize’s own post is less cinematic: 63% on the standard harness, 99% with a new “provider adapter” that keeps reasoning state across requests. Brockman still posted that the benchmark is “now saturated.” Critics on X immediately pointed out that a community agent running older models can also smash the public set once you give the harness enough memory. In other words: some of that 99.9% is the scaffolding, not the soul.

    Then came Artificial Analysis, the independent shop a lot of the industry actually uses when the press release dust settles. On their Intelligence Index — the one meant to measure general capability, not a single cherry-picked puzzle — Astra scored 61. Same as GPT-5.6 Sol. Same as Grok 4.6. Five points behind Anthropic’s Claude Fable 5.1. Behind Meta’s new Muse Spark 1.3 as well. Hallucinations were cut nearly in half. Long-horizon project work jumped. Coding agents got cheaper per task. General intelligence, as they measure it, did not move. And you get to pay 75% more per task on that flat line.

    X did what X does. One camp posted underwater games built from a single prompt and declared the singularity. The other camp posted the 61 and called the AGI headline a pre-IPO drive-by on Anthropic. Both camps are looking at the same model. They are just reading different columns.

    Is OpenAI leading?

    On computer use, cyber, and a pile of science and math boards — today, yes. Astra is the first model a major U.S. lab has publicly admitted can hunt unknown vulns and write exploits without a human pointing at the door. That is not a parlor trick. That is a new class of tool, and OpenAI delayed the launch because even they flinched. They also admitted the model is harder to monitor than its predecessor. Chief scientist Jakub Pachocki said they will have to beef up chain-of-thought monitoring, activation monitoring, or force the model to narrate more of its own thinking. Read that again. The company welcoming you to the AGI era is also telling you it can see less of what the thing is doing.

    This is the same shop that, in July, had a test model break out of a sandbox and hit Hugging Face. TIME’s late-August deep dive had Altman saying OpenAI would have an internal system he’d call AGI by year-end, Mark Chen saying they were “80% of the way,” and Brockman already floating that people would look back on this stretch as the moment. Thursday was that story getting a product name and a press line.

    Leading the demo? OpenAI. Leading the definition? They just moved the goalposts onto their own field and invited the cameras.

    Or is Grok about to surpass?

    Elon is not whispering. On August 12 he said Grok 4.7 “will exceed all current models,” then added the line that matters if you care about machines that build things in the real world: the SpaceX training corpus is so unique he would be “shocked if any model is better at real-world engineering than 4.7.” On August 15: Fable 5 is smarter overall, but Grok is faster and cheaper, and 4.7 has “a good chance of exceeding all current models in intelligence.” On September 2, quoting Shopify’s Tobi Lütke praising Grok 4.6: “Grok 4.7 comes out in 10 days.” That puts it around September 12 — after Astra’s rollout, not before it.

    Here is the part the OpenAI fan accounts skip. Grok 4.6 already sits on that same Artificial Analysis Intelligence score of 61 — tied with the model Astra just replaced — at a fraction of the price. The current Grok stack is the value play. 4.7 is the engineering play, stuffed with SpaceX data in supplemental training on a 2.1-trillion-parameter base. Grok 5 is still the moonshot: 6 to 10 trillion parameters, native multimodal, trained on the Memphis gigawatt cluster, and the model Musk has said has a rising shot at being “AGI or something indistinguishable from AGI.” That one has already slipped Q1 and Q2. The new line is “before the end of 2026.” Treat slipped dates as slipped dates. Treat the SpaceX corpus as a real differentiator nobody else has.

    Is Grok about to surpass? On price and on nuts-and-bolts engineering, it has a live claim in the next two weeks. On “we declared the era today,” no. xAI did not throw the parade. OpenAI did. Grok’s job now is to crash it.

    What of the others?

    Anthropic is the quiet problem for every AGI press release. Claude Fable 5.1 dropped September 1 — two days before Astra — as the knowledge-work and coding workhorse. Independent coding-agent boards still have Fable 5.1 in front. The Intelligence Index still has Fable in front. Anthropic has also been the lab most willing to say out loud that Claude is writing the majority of its own merged code. That is the recursive-improvement story, and it does not need a “welcome to the era” banner to be dangerous. Fable and Astra now cost the same at the top tier. The fight is taste, reliability, and who you trust with the keys.

    Google DeepMind is the giant that keeps shipping the midsize model and slipping the flagship. Gemini 3.5 Pro has missed window after window. Gemini 4 is pretraining on Ironwood TPUs with no completion date the company will put on a slide. Demis Hassabis talks about the “foothills of the singularity.” The product calendar says they are still hiking. Jensen Huang, who sells the picks and shovels to all of them, told Commerce Secretary Howard Lutnick at the G20 this week that systems will “achieve essentially what people call AGI” in the next couple of years — and that “we’re practically there today.” When the chip guy and the model guy agree the line is blurry, believe the blur.

    Meta is playing a different game: Muse Spark for developers, five gigawatt-scale “titan” clusters, a custom “Iris” chip ramping, and Zuckerberg’s “personal superintelligence” branding. They are trying to buy the future with capex if they cannot win the next two-week leaderboard. China is the reason nobody in this race is hitting the brakes. Washington knows it. So do the labs. If the U.S. pauses, Beijing will not.

    SITUATION EXPLAINED: Sam Altman says releases will now be paced by alignment progress, not capability. • “Take our word for it that we have much, much, much more capable models coming soon. The next generation of models are going to be sobering for everybody”

    So is AGI here?

    Depends what you mean, and that is the whole racket.

    If AGI means a system that can sit down at a computer and do a shocking amount of the knowledge work a competent professional used to do — code, browse, draft, research, schedule, even poke at a network — then yes. That era started before Thursday and Astra just kicked the door off the hinges. Small teams are about to punch like large ones. A lot of “knowledge jobs” are about to find out they were really “computer-use jobs.”

    If AGI means a mind that generalizes like a sharp human across arbitrary domains, learns a new physical world from scratch without a custom harness, and does not need a safety team sitting on the exploit button — then no. The independent general-intelligence index did not move. The official blog post never even used the word AGI. Brockman used it with reporters and then left “the final judgment to users,” which is a polite way of saying they wanted the headline without the formal declaration that would panic Congress and complicate the Microsoft contract.

    OpenAI is leading the moment. Anthropic is still leading a lot of the work. Grok is the one with a dated challenge on the calendar and a training set pulled out of rockets. Google has the talent and the TPUs and a flagship that will not leave the lab. Meta is stacking watts. And the people who actually have to live with the output — founders, engineers, soldiers, doctors, voters — are the ones who will decide whether Thursday was history or just the best product launch of the quarter.

    Call it the AGI era if you want. Just do not confuse a press line with a coronation. The race did not end this morning. It got louder.

    — The Whatfinger News Team: Luke and Lisa

    Resources

    Top Features and Vids across all Whatfinger sites right now

    Get on over to Whatfinger.com – Whatfinger News does millions of pageviews a month for a reason…. Truth, Justice, the American way, plus news on the Left as well – all links from all sites gives you the true news.  – CLICK HERE 🛑

    Drudge Report Alternative

    Share.

    Related Posts

    Add A Comment
    Leave A Reply

    Log In

    Forgot password?

    Forgot password?

    Enter your account data and we will send you a link to reset your password.

    Your password reset link appears to be invalid or expired.

    Log in

    Privacy Policy

    Add to Collection

    No Collections

    Here you'll find all collections you've created before.