Claude Opus 4.8 low-cost reasoning mode sometimes outperforms prior max tier
Claude Opus 4.8 ships with a new low-cost reasoning mode that sometimes outperforms the prior max tier, while the fast variant drops to one-third its previous price.
Anthropic released Claude Opus 4.8 this week with a restructured pricing ladder: a new low-cost reasoning mode that occasionally outperforms the previous max-tier configuration, and a fast variant priced at roughly one-third its prior cost.
The model introduces two structural changes. A "low" reasoning mode trades compute for accuracy—yet in some tasks surpasses the older max setting—and a fast variant now priced at one-third the cost of its predecessor. Anthropic describes Opus 4.8 as "more honest," meaning it avoids shortcut reasoning and admits uncertainty more readily than earlier versions.
The pricing shift arrives as competition intensifies in the hosted LLM market. Opus 4.8's fast tier now undercuts several rivals on cost-per-token while maintaining comparable output quality on standard benchmarks. The low mode offers a middle ground for workflows that can tolerate slightly longer inference in exchange for better reasoning fidelity.
Anthropic has not yet published a formal technical report or benchmark suite for Opus 4.8. The model is available now via the Claude API, with the fast tier accessible to all existing customers and the low mode rolling out over the next week.
