Anthropic Launches Claude Opus 5, Priced for Agentic Workflows as Enterprise AI War Heats Up

Anthropic made Claude Opus 5 available to developers and enterprise clients on August 1, positioning the model around "cost per accepted task" — an efficiency metric for autonomous agent workflows — as the competitive battle with OpenAI shifts from raw benchmark scores to agentic economics. According to a Chinese brokerage research note reported by Sina Finance, Opus 5 significantly outperforms its predecessor on coding, knowledge work, and life sciences benchmarks, with intelligence approaching Anthropic's flagship Fable 5 at roughly half the price.

Pricing the Agent Economy

The launch lands as Anthropic presses an enterprise advantage: Blackstone recently selected Anthropic over OpenAI as its AI tools partner, and Claude Code has been steadily absorbing developers from OpenAI's Codex. OpenAI responded this week by cutting prices on GPT-5.6 Luna by 80% and Terra by 20% while raising usage allowances — a sign that both labs now compete on per-task economics rather than headline model quality alone.

"Cost per accepted task is becoming the unit of competition in agentic AI," the brokerage note argued — a framing that favors models which complete multi-step work reliably on the first attempt.

A Crowded August

The release follows a record July of model launches, including Google DeepMind's Gemini Omni Flash and Gemini 3.6 Flash, and precedes an expected autumn IPO window for Anthropic, whose May funding round valued it at $965 billion — above OpenAI's $852 billion March valuation, per this week's industry wrap.