Back to AIBriefs
AnalysisAI Models

Opus 4.8 Thinking consumes up to 900K cache tokens per turn

User reports Opus 4.8 with Thinking writes up to 900,000 cache tokens per turn, vs Opus 4.7's 14,000–34,000. Context snowballs, draining windows in minutes instead of hours.

·
14 days ago
Opus 4.8 Thinking consumes up to 900K cache tokens per turn — AIBriefs