Claude Sonnet 5 Ships at $2/$10 — and Runs Agents Like Opus
Anthropic just made near-Opus agentic capability the default at 60% below Opus pricing. Cheaper tokens are a real lever — but only if your pipeline isn't quietly multiplying them.
NeuroX AI · July 3, 2026
Anthropic shipped Claude Sonnet 5 on June 30 and made it the default model for every Free and Pro user the next day. The headline isn't the benchmark — it's the invoice. Sonnet 5 lists at $2 per million input tokens and $10 per million output through August 31 ($3/$15 after), against Opus 4.8's $5/$25. That's roughly 60% cheaper, at what Anthropic positions as near-Opus agentic quality.
The timing is deliberate. Q2 was the quarter enterprises recoiled from agentic bills — long autonomous runs that read dozens of files, retry their own errors, and burn a token budget sized for a quarter in a couple of weeks. When an agent runs for hours, the per-token price is the cost model. Halving it changes what you can afford to put into production.
But a cheaper token doesn't fix a leaky loop. An agent that re-reads context it already has, retries without backoff, or skips prompt caching just multiplies a smaller number by a bigger one. We've watched a model swap "save" 60% on paper and net nothing, because the harness was the real spend.
Model price is one lever. Token efficiency is the other — and it's the one a demo never shows.