Anthropic launches Claude Opus 5 at half the price of Fable 5

Announced July 24 and covered widely through the week, Claude Opus 5 is Anthropic's play in the intensifying price war. At $5/$25 per million input/output tokens it undercuts the top-tier Fable 5 by half while approaching its capabilities on coding, knowledge and agentic tasks. Cost-per-task analysis circulating among developers pegged Opus 5 at roughly $2.03 versus $2.75 for Fable 5, a comparison that resonated with enterprise and indie developers optimizing API spend.

This is Anthropic's fourth Claude 5 release in under two months, a cadence that has itself become a talking point: observers note AI has shifted from blockbuster launches to incremental cost-and-speed improvements, with Opus 5 pitched squarely as a cheaper everyday assistant rather than a capability leap. Community sentiment was polarized — a 1,251-upvote demo of an NMS-style exploration game built in a day with Opus 5 contrasted with threads like 'Claude Opus 5 is an asshole' (284) and 'Let me know when Anthropic releases a good model again' (226).

Independent benchmarks tempered the hype. LlamaIndex's Jerry Liu found Opus 5 'roughly on par with Opus 4.8' on document understanding — slightly worse on dense tables, slightly better on charts — and advised against using it for large-scale document parsing at 8 cents per page, while praising it for coding and knowledge work.

Around the launch, Anthropic also declined to sign Jensen Huang's open-model letter (though it joined a petition urging the U.S. to 'deliberately pace' frontier AI), and its $1.5 billion copyright settlement received final approval. The positioning is consistent: aggressive on price and everyday utility, cautious and closed on weights — a stance that drew fire during the same week's security and open-weights debates.