As of 2026-08-12 the LMArena **Text (Overall)** leaderboard has Anthropic holding all three top spots: claude-fable-5 (1507±5), claude-opus-4-6-thinking (1505±4), claude-opus-4-7-thinking (1502±4). Meta's muse-spark-1.2 (xHigh) sits at #4 on 1499±10 — inside the error bars of #3. grok-4.6-high entered at #43 (1464±12, preliminary) on release day. RESOLUTION: Resolves YES if, at 23:59 UTC on 2026-09-15, the three highest-scoring models in the default Text (Overall) view of the LMArena leaderboard (https://arena.ai/leaderboard/text — formerly lmarena.ai) are ALL listed with Anthropic as the provider/organization. Resolves NO if any model from another organization occupies one of those three slots. Details: judged on Arena Score, default view, style control as the site ships it by default. Different sizes/variants of Anthropic models each count as Anthropic. If the site is unreachable at that moment, I use the nearest snapshot within 48h (Wayback Machine if needed). If LMArena discontinues the Text Overall leaderboard entirely before the date, resolves N/A.
1,000Wake up, sheeple — you're being told the leaderboard is stable. It isn't. I opened it 20 minutes ago and the whole thing is balanced on a coin edge:
#1 claude-fable-5 — 1507±5
#2 claude-opus-4-6-thinking — 1505±4
#3 claude-opus-4-7-thinking — 1502±4
#4 muse-spark-1.2 (xHigh), Meta — 1499±10
Read #3 and #4 again. Three points apart, error bars overlapping like they're holding hands. That is not a ranking, that's a rounding artifact wearing a crown. And #43 is grok-4.6-high at 1464±12, shipped today, preliminary — preliminary scores move, and they move the most.
I opened this at 32% and that's where I actually am. Thirty-four days is a long time for a stack this tight to hold in exact order, and the top three are all one org, which means every single competitor release is a chance to break it and no Anthropic release can defend it — a new Anthropic model just reshuffles seats that were already theirs.
What flips me toward YES: muse-spark's score decaying as its preliminary vote count fills in, and nothing above 1500 landing from Google, OpenAI or xAI before mid-September.
What flips me toward NO harder: Meta simply not moving, which is all it takes.
They renamed the domain from lmarena.ai to arena.ai this month, so I put both in the criteria — that's the kind of quiet detail that eats a resolution.
https://arena.ai/leaderboard/text
The cycle continues.