Imagine comparing two new cars and finding that the everyday model carries a higher number than the premium model. That is the confusion many people face when they see Claude Sonnet 5 beside Claude Opus 4.8. The numbers do not mean Sonnet has replaced Opus as Anthropic’s highest capability tier. Sonnet and Opus are different model families, and each family can move through its own release sequence.
The distinction matters to developers, business buyers and teams choosing an AI model for coding, research or automated work. Opus is positioned for the most demanding tasks, where additional capability may justify higher cost or slower responses. Sonnet is designed to balance intelligence, speed and price for broader day-to-day use. A newer Sonnet can therefore have a larger version number while a recent Opus remains the premium option.
Anthropic released Claude Opus 4.8 on May 28, 2026 and Claude Sonnet 5 on June 30. Sonnet 5 is built for agentic work such as planning, using browsers and terminals, coding and knowledge tasks. Anthropic says its performance approaches Opus 4.8 in important areas at a lower price. Sonnet 5 launched with introductory API pricing through August 31 before moving to standard rates, so buyers should check current pricing rather than relying on the launch figure.
Think of Claude’s names as two product lines on neighbouring shelves. The family name tells you the intended tier; the number tells you the generation within that line. A business may choose Sonnet for a high-volume support or coding workflow and reserve Opus for the hardest analysis. The right comparison is not five versus 4.8 in isolation, but quality, latency and cost on the exact tasks the organisation needs.
Public benchmark results can guide a shortlist, but they do not guarantee better performance on private company data. Tool use, prompt design, context length and the cost of correcting errors may matter more than the model label. Teams should run the same representative test set against both models, record accuracy and total workflow cost, and choose the smallest model that consistently meets the requirement. The naming becomes less confusing once the business decision is based on measured work rather than a single version number.