Claude Opus 4.8: what it is, what it costs, where it wins
Opus is Anthropic's heaviest model, the one Claude.ai reserves for its $20+ plans. It reads whole codebases and 200-page contracts without losing the plot, and it is the model most often named best writer in blind tests. Below: what it does better, what it costs everywhere, and how to try it in the next 60 seconds without a subscription.
Where it wins
Three tasks where Opus 4.8 consistently beat the field in our July test set: long-document analysis (a 74-page lease, zero hallucinated clauses), nuanced editing that preserves the author's voice, and multi-step reasoning over messy spreadsheets. For quick factual questions it is overkill, a 30x token burn for answers Haiku gives at 6x.
What it costs, honestly
On Claude.ai, Opus sits behind a $20/mo plan with opaque usage caps. Via API it is $5/$25 per million tokens. Here it is an Ultra-class model: a Plus budget buys 25 to 45 Opus messages a month, and the price of each message is printed next to the send button before you commit.
Where it loses
Speed: first tokens arrive noticeably later than Sonnet's. Cost discipline: it happily writes 900 words when 90 would do, which matters at 30x. And for code generation specifically, DeepSeek v4-pro matched it on 34 of 40 tasks at a fraction of the burn. Switching mid-dialog is the honest fix: draft with Opus, iterate with something cheaper.
Fact sheet
Guest messages run on economy class.