Opus 5.5 tops writing benchmark at 2631 Elo, 307-point lead; max tier costs $3.43 per script
OnlyProggingForFun · reddit · 2026-09-27
An ongoing AI writing benchmark (167 model configs; 10 script tasks × 5 scripts, blind-scored by AI judges from three labs) reports Claude Opus 5.5 debuting at #1 with 2631 Elo — a 307-point gap over second-place Fable (2324), the largest single jump since the board launched in June 2026, and the first model to clear 91/100 on its rubrics.
Effort-tier breakdown:
- max: #1, 2600 Elo, but $3.43 and 17 minutes per script — slowest config on the entire board by far
- xhigh: #2, 2399 Elo, $0.86, 4.4 min — beats the previous champion by 100 Elo at 1/4 the price
- high: #3, 2342 Elo, $0.34, 1.7 min — also beats it at 1/9 the price
- medium: 2199 Elo at $0.21; low: 2144 Elo at $0.17
- With no effort flag passed from a Claude subagent, the model self-selects and lands between medium and high
Author's take: high is the sweet spot for most writing; use max when you want the best draft and can wait. Previous champion Fable 5.1 max sits at 2303 Elo, $3.15/script.
More from Models
- Ethan Mollick: Opus 4.7-5 lost the 'Claude feel', Opus 5.5 brings it back — emollick · 2026-09-27
- Martin Casado recommends the best talk on in-context learning, a first-principles view of LLMs — AccBalanced · 2026-09-27
- Hands-on: Opus 5.5 high beats GPT-6 astra xhigh on real Pagespeed optimization — mazzaTalk · 2026-09-27
- Dev on Opus 5.5: smooth multi-part coordination flips coding dynamic — ezshine · 2026-09-27
- Asked Grok to teach Japanese kanji, it started inventing its own — JoeJustice · 2026-09-27
- Melanie Mitchell backs claim that today's AI 'are not LLMs' anymore — asusarla · 2026-09-27