Meta's first frontier models score 52 and 56 against 58

Meta released Muse Wonder and Muse Wonder Full, its first frontier models, and Artificial Analysis scored them at 52 and 56 on its Intelligence Index against 58 for Claude Opus 5.5 (TechCrunch).

That is the measured distance on today's record between a company's first frontier release and the current best: six points across the band, two points at the top of it. Every other number on the record is denominated in something much larger.

The commitments are in gigawatts

Microsoft said it will install its own Maia 300 accelerators in its data centers next year, with Maia deployments the company expects to be measured in multiple gigawatts (Bloomberg).

Set that beside the index. The capability difference between the newest entrant and the leader is a two-point gap on a benchmark. The infrastructure being committed to hold a position in that band is stated in the unit a national grid is measured in, and committed a year ahead of the silicon arriving.

The market moved on the hardware, not the scores

South Korea's KOSPI fell 7.4% to 5,914.35, its lowest since late July, as Samsung and SK Hynix led a regional chip selloff that also took China's CSI 300 to a 14-month low and the Nikkei down about 2.2% (NBC News).

Nothing on today's record says a model got worse. What repriced is the layer underneath: the companies that sell the parts, marked down 7.4% in Seoul and to a 14-month low in China on the same day a hyperscaler said it intends to stop buying so many of them. Maia 300 is in-house silicon. A company installing its own accelerators at gigawatt scale is a customer the firms that fell today are planning to lose.

What sits behind the top score

The 58 that Meta's two models are measured against belongs to Anthropic, which the ledger records raising $65B in Series H at a $965B post-money valuation in May, passing OpenAI as the most valuable AI startup for the first time (Anthropic).

That is the shape of the asymmetry. Six index points separate the whole visible field, and the capital, the power contracts and the market caps arranged around those six points are not six percent apart. The index is the one quantity on today's record that is converging. Nothing attached to it is.

Brussels is writing the procedure, not the threshold

The European Commission opened a four-week consultation on a draft implementing regulation for AI Act enforcement under Articles 75c and 75d, covering access to the file, procedural safeguards and limitation periods for penalties, with feedback closing on November 9 and adoption planned for the fourth quarter (PPC Land).

Read what the consultation is actually about. Not capability thresholds, not evaluation standards, not what a model has to score. Access to the file, procedural safeguards, limitation periods: that is the machinery a regulator drafts when it expects to run proceedings against named providers and wants the penalties to survive appeal. The enforcement apparatus is being specified at the same time as the field it applies to is compressing into a six-point band.

Four items, four units

Two index points at the top of the field. Multiple gigawatts, committed for next year. A 7.4% single-day fall in Seoul. A four-week comment period closing November 9.

Only the first of those measures what a model does. The other three measure what has been built, bought and drafted around the assumption that the differences between models are larger than the benchmark says they are. On today's record, the scores are the narrowest thing in the frame.