For months, building an autonomous AI agent meant paying flagship prices. Anthropic changed that on June 30 with Claude Sonnet 5 — a mid-tier model that can plan tasks, browse the web, execute terminal commands, and operate autonomously at a level that previously required Opus-class compute.
The benchmark gap is narrow. Sonnet 5 scores 63.2% on agentic coding, versus Opus 4.8's 69.2% — a six-point difference. The price gap is far wider. Opus 4.8 costs $5 per million input tokens and $25 per million output tokens. Sonnet 5 starts at $2 and $10, a 60% discount. That introductory rate holds until August 31, after which pricing moves to $3 and $15. For context, Sonnet 4.6 scored just 58.1% on the same benchmark — Sonnet 5 jumped five points in one release.
Safety also improved. Compared to Sonnet 4.6, the new model shows a meaningfully lower rate of undesirable behaviors — better at refusing misuse requests and more resistant to prompt-injection attacks, where malicious instructions are embedded in the data an agent processes. For long-running autonomous workflows with minimal human oversight, that distinction matters.
The timing is not accidental. Anthropic is racing toward its IPO, and a cheaper high-performance option widens the developer market considerably. Startups that could not justify Opus pricing now have a compelling alternative. The question is whether OpenAI and Google respond with similar pricing cuts, or cede the cost-sensitive segment to Anthropic.
