TechOpus 5.5 reportedly beats Opus 5 on ARC-AGI at roughly 80% lower evaluation cost
ARC Prize lists verified scores for Anthropic’s Opus 5.5: 93.3% on ARC-AGI-2 at $0.41 per task and 98.5% on ARC-AGI-1 at $0.16 per task.
TLDR
ARC Prize says Opus 5.5 outscored Opus 5 by 2.9 percentage points on ARC-AGI-2 and 1.0 percentage point on ARC-AGI-1, at roughly 80% lower evaluation cost. It labels the results verified.
Separately, a user says switching effort mid-session on Opus 5.5 preserves the prompt cache, provided you’re using Claude Code v2.1.280 or later.
Opus 5.5 reportedly beats Opus 5 on ARC-AGI at roughly 80% lower evaluation cost
ARC Prize lists verified scores for Anthropic’s Opus 5.5: 93.3% on ARC-AGI-2 at $0.41 per task and 98.5% on ARC-AGI-1 at $0.16 per task.
