Google Takes the AI Crown, and the Benchmark People Won't Stop Refreshing

A 0.8% longshot lands: the Bureau of Vibes and Evals ranks one model above every rival at the July 31 bell, and nobody quite believes the scoreboard.

The scoreboard blinked over at 11:58 pm on July 31, 2026, and it stayed put. Google held the top slot on every eval the Bureau of Vibes and Evals runs, ending a chase the market had priced at basically zero.

"We reran it four times because the ranking felt like a typo," said Dr. Ines Okafor-Halvorsen, chief vibes officer at the Bureau. "It was not a typo."

The mechanism was quiet. A late model update shaved reasoning latency by 41 percent while lifting the Bureau's long-context accuracy from 88 to 96, and the gains held across the whole eval suite instead of one cherry-picked task. That combination is what flipped the standings. Rivals held steady. Google simply did not stop climbing.

Halcyon Compute, the newcomer everyone expected to spoil the party, finished fourth. On the trading floor the ripple was immediate, with one desk taping the leaderboard to a monitor at 12:02 am and refusing to take it down.

"Zero-point-eight percent outcomes are why we keep the popcorn stocked," said Dr. Casimir Blum of the Meridian Macro Institute. The same night oil dipped to 65 and Bitcoin cleared 82,500, so the improbable had company.

Verification runs continue Monday.

— Timeline B. In our universe, the odds of this were 0.8%.