Anthropic Published Two Different Scores for Claude Sonnet 5.5. Here's Which Is Right
Claude Sonnet 5.5 Terminal-Bench 4.0: Anthropic reports 70.6%, Artificial Analysis measures 63.6%. What the 7-point gap means and which number to use.
AI benchmarks taken apart: who ran it, how it was scored, and what the number means.
AI engineers, researchers, model and AI companies; not mass market.
Opus 5.5 AutomationBench Score: Why Two Labs Report Different Numbers for the Same Model
Opus 5.5 AutomationBench score is 40.0% in Anthropic's launch table and 42.47% on Zapier's leaderboard. The gap comes from how fallbacks are scored.
Terminal-Bench 4.0 Explained: What Changed From 2.1 and 3.0, and Why Scores Do Not Carry Over
Terminal-Bench 4.0 removes 8 tasks, fixes 19 and sets an 8-hour timeout. Opus 5.5 scores 66.4% (lab-reported) vs 59.6% independent. Why versions don't compare.
Gemini 4 Argon vs GPT-6 Astra: Same Score, but One Hallucinates 3x More
Gemini 4 Argon and GPT-6 Astra tie at 53 on the Artificial Analysis index, but The Batch reports AA-Omniscience hallucination of 15% vs 51%.
Gemini 4 Argon Cost per Task: The Hidden Token Bill Benchmarks Don't Show You
Gemini 4 Argon cost per task: Artificial Analysis measures $1.99 at promo pricing and projects $3.98 at list, against $3.26 for GPT-6 Astra.
ARC-AGI-3 Was "Beaten" at 99.9%. Change the Harness and It Drops to 62.7%
ARC-AGI-3 harness gap explained: ARC Prize measured GPT-6 Astra at 62.7% with its Standard harness and 99.9% with a Provider Adapter. What changed.
Claude Opus 5.5 on Terminal-Bench 4.0: Anthropic Says 66.4%, Independent Test Says 59.6%
Claude Opus 5.5 Terminal-Bench 4.0: Anthropic reports 66.4%, Artificial Analysis measures 59.6%. Why a 6.8 point gap exists and how far to trust either.