The rival scoreboards joined the corpus today — and they can't agree on who's winning
The Aggregate Digest — Saturday, August 1, 2026
The rival scoreboards joined the corpus today — and they can't agree on who's winning. Two composite indices published by other aggregators are now tracked as boards in their own right. LM Market Cap's LMC Score (342 models) crowns Claude Fable 5 at 97.1, with a wall of Claude Opus rows behind it — Opus 5, 4.8 and 4.7, each in standard and "Fast" serving — all tied at 95.1. LLM Stats' conservative rating (320 models) sees it differently: GPT-5.6 Sol on top at 58.05, Claude Opus 5 just 0.07 behind at 57.98, and Claude Fable 5 third at 57.51. Our own fused ranking splits the difference: Claude Mythos 5 leads the model field, with Claude Opus 5, GPT-5.5 Pro, GPT-5.6 Sol and Claude Fable 5 packed within a ten-point ELO band behind it. Three scoreboards, three different names at the top — a live illustration of why single-index rankings are worth tracking side by side with the benchmarks they summarize.
One genuinely new model arrived. Celeris-1 debuts mid-pack: provisional ELO 1518, #632 of 1,484 on the full board, off ten fresh scores — 63.1 on AA GPQA Diamond (326th of 555), 27.0 on AA Long Context Reasoning, 6.6 on AA Humanity's Last Exam, and an Artificial Analysis Intelligence Index of 11.8 (337th of 554). A debut, not a splash — but ten boards on day one is enough for a real read.
Elsewhere. GPT-5.6 Sol took third on Design Arena (SVG) at 1341 Elo, slotting in behind Claude Fable 5 (1349) and prism (1345) — so the model LLM Stats calls the world's best is drawing SVGs slightly worse than the model LM Market Cap prefers. And on MERA, the Russian-language evaluation suite, Claude Opus 4.6's 0.86 sits a hair above the board's own human benchmark at 0.85 — one of the quieter human-baseline crossings on record.
