53 to 53. One Of Them Is Wrong Half The Time.

Two models walked into a leaderboard. They tied. That is the whole story if you only read the press release.

With high reasoning, Gemini 4 Argon scores 53 on the Artificial Analysis Intelligence Index, matching GPT-6 Astra at 53 and edging GPT-6.1 Sol at 52. Google is already calling this a comeback. It is not wrong, exactly. It is just not the number that matters.

The number that matters is this: on AA-Omniscience, a benchmark that tests factual knowledge and honest handling of knowledge gaps, Argon's hallucination rate is 15 percent, while GPT-6 Astra comes in at 51 percent and GPT-6.1 Sol at 54 percent. Same intelligence score. One model is wrong, confidently, roughly every other time it opens its mouth.

The tie is real. The equivalence is not.

Argon is far more likely to admit it doesn't know an answer rather than guess wrong, but its accuracy reaches only 50 percent, 13 points below Astra's 63 percent. Astra guesses more and gets more right when it guesses. Different failure modes, not one model being simply better. A customer-support bot that says "I don't know" constantly is annoying. A customer-support bot that's confidently wrong half the time is a lawsuit.

That asymmetry is exactly why the Intelligence Index tie is theater. It collapses two completely different risk profiles into one number that goes in the sales deck.

At current discounted pricing, Gemini 4 Argon costs $1.99 per Intelligence Index task, 60% of GPT-6 Astra's $3.26. Cheaper and more honest about its own ignorance. Also less accurate when it does commit to an answer. If your procurement team is buying off the top-line score alone, they're buying blind on the one axis that determines whether your product embarrasses you in public.

Washington's answer: change the label

While labs argued over decimal points, the White House solved a different problem entirely. President Donald Trump ordered all executive branch departments and agencies to use the term "super intelligence" instead of artificial intelligence.

Translation: stop measuring how often we're wrong, start admiring how smart we sound.

The timing is not subtle. The rebrand comes weeks before the midterm election, which appears increasingly shaped by Americans' concerns about rapid AI advancement. You don't rename a category unless the old name has become a liability. "Artificial" implies fake. "Super intelligence" implies trustworthy. Neither word has anything to do with whether the model answering your tax question is right.

Nobody in this story fixed the 51 percent hallucination rate. Google shipped a model that abstains more. OpenAI shipped Astra months ago and hasn't addressed the gap publicly. The administration renamed the discomfort. Three different responses to the same underlying fact, and only one of them is actually a fix.

If you're the enterprise buyer choosing between Argon and Astra for anything that talks to a customer, the leaderboard tie tells you nothing. The hallucination rate and the accuracy number sitting next to it tell you what you're actually signing up for.

What to watch: whether Artificial Analysis treats hallucination rate as a co-equal metric to the Intelligence Index instead of a footnote. Whether OpenAI responds to a now-public 51 percent figure with anything beyond a blog post. And whether any federal agency actually adopts "SI" in a document that matters, versus quietly ignoring an order with no enforcement teeth.