Anthropic just dropped Claude Opus 5, and the math is hard to argue with. The model costs $5 per million input tokens and $25 per million output tokens, exactly half the price of the company’s higher-tier Fable 5. It also beats Fable 5 on most major benchmarks.
What Opus 5 actually does differently
Opus 5 uses approximately one-seventh the reasoning tokens of its predecessor, Opus 4.8. It also runs at less than half the latency.
Anthropic also introduced a new “effort” setting that lets users prioritize between intelligence, token usage, and processing speed. For agentic tasks, the kind where AI operates semi-autonomously across multiple steps, this flexibility matters enormously.
Early user feedback has been strong across software engineering, data analysis, and what Anthropic calls “complex agentic tasks.” The model posted top scores within the Opus family on proprietary trading benchmarks.
