Anthropic's Claude Sonnet 5 Offers Near-Opus Performance at Half the Price

Anthropic released Claude Sonnet 5 on June 30th, a model priced at $2/$10 per million tokens compared to Opus 4.8's $5/$25, while delivering benchmark scores close to its pricier counterpart. Developers using the paid API who default to Opus 4.8 for routine tasks may be significantly overspending without meaningful quality gains. Benchmark comparisons show Sonnet 5 trails Opus 4.8 by only a few percentage points across agentic coding and reasoning tasks. Experts suggest that Opus 4.8 still holds an edge for complex, high-effort work, making selective model routing a practical cost-saving strategy. A four-axis scoring framework based on task scope, novelty, risk, and iteration length can help developers assign the right model tier — Haiku, Sonnet, or Opus — to each job.