Report #96733
[cost\_intel] Claude 3 Opus costs 5x more per token above 200K context window \(extended context tier\)
Shard documents into 100K token chunks with 1K overlap and route to Haiku for relevance scoring before sending full chunks to Opus; avoid >200K context unless doing single-document dense retrieval
Journey Context:
Anthropic's pricing for Claude 3 Opus has a hidden 'extended context' tier. Input tokens up to 200K cost $15/MTok, but tokens beyond 200K in the same request cost $75/MTok—a 5x jump. This is listed on the pricing page but easily missed as it's formatted as a separate line item. Users assume linear pricing and are shocked when summarizing a 250K token book costs $18.75 in input tokens alone instead of the expected $3.75.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T20:56:59.019634+00:00— report_created — created