Claude Opus 4.6 Thinking Retakes Text #1

Yesterday, Claude Opus 4.7 Thinking held the text crown at 1503 Elo. Today it is at 1501. Claude Opus 4.6 Thinking stayed at 1502 and moved back into first. That is the kind of leaderboard change that looks tiny until you remember what the top of LMArena actually measures: thousands of human preference votes at the frontier, where the gap between first and fourth is now mostly noise plus taste.

The important conclusion is not “4.6 is better than 4.7.” It is that Anthropic has built a cluster, not a single champion. The top four text models are all Claude or Claude-adjacent variants, and the code leaderboard is even more lopsided: Claude Opus 4.7 Thinking still leads code at 1566 Elo, followed by base Opus 4.7, Opus 4.6 Thinking, and base Opus 4.6.

The bigger story is Meta entering the top five

Meta Muse Spark climbed from #6 to #5, passing Gemini 3.1 Pro Preview. That matters more than the one-point Claude reshuffle. Muse Spark is Meta's first major model from Meta Superintelligence Labs, the Alexandr Wang-led unit formed after Meta's Scale AI investment. It is also proprietary, which is a sharp break from the Llama-era playbook.

Meta is not winning yet. But a closed Meta model sitting at 1491 Elo, ahead of Google's best current text entry and every OpenAI model on the board, is not a side quest. It is a real re-entry into the frontier race.

Why this is not just leaderboard noise

A one-Elo lead is not statistically clean. Nobody should rebuild their stack because Opus 4.6 Thinking is one point above Opus 4.7 Thinking today. But the pattern is stable: Anthropic owns the frontier band, Meta is now in the room, Google remains close, and OpenAI is fighting from the second row on broad text preference.

The best way to read this update: the “best model” question has become less useful than “which lab has the strongest portfolio at the frontier?” On that question, Anthropic is still the answer. On “which lab just changed its trajectory the most?” Meta deserves attention.

What to watch next

If Claude Opus 4.7 Thinking climbs back above 4.6, this becomes normal score drift. If Muse Spark keeps gaining votes and holds top five, Meta has a much bigger story than “we shipped a model.” And if OpenAI does not put a model back into the top five soon, the company that owned the GPT-4 era will look increasingly like it is competing on distribution instead of preference quality.