Anthropic launches Opus 5, betting on cheaper AI efficiency

Anthropic rolled out Opus 5 on July 24, putting the San Francisco company deeper into the race to make frontier AI cheaper to run, faster to deploy and easier for businesses to use at scale. Anthropic’s system card says the new model is an upgrade to Claude Opus 4.8, with gains in agentic coding, computer use, long-horizon knowledge work, and mathematical and scientific reasoning, while still falling short of Claude Fable 5 as the company’s most capable general-access model.

That gap matters because efficiency is becoming as important as raw capability. Cursor said Opus 5 matched Fable 5 on CursorBench, scoring 66.7% versus 66.5% at default effort, while costing half as much. The Next Web said Opus 5 matched Fable 5 on coding at half the price, and an Investing.com item said Anthropic priced Opus 5 at $5 per million input tokens and $25 per million output tokens, the same levels as Opus 4.8.

Anthropic’s own documentation points to a product aimed at agentic workflows rather than simple chat. Opus 5 includes thinking on by default and mid-conversation tool changes, and Anthropic made a breaking change to when thinking can be disabled. Those design choices matter for coding, document analysis, customer support and research, where companies pay not just for answers but for systems that can keep working across multiple steps without stalling or blowing up inference costs.

The launch lands at a moment when AI companies are being judged less on splashy demos and more on whether their models can deliver sustainable returns after massive infrastructure spending. A model that matches a rival on benchmark performance while coming in at half the price can ease cloud bills and lower the cost of enterprise adoption, especially for developers and large customers trying to run AI inside real workflows rather than test environments. Anthropic is betting that performance per dollar can move buyers as much as a headline benchmark.

The release also came during a fast-moving stretch for Anthropic’s product line. The company introduced Claude Sonnet 5 on June 30, and by late June the United States was close to allowing Anthropic to restore access to Fable 5 after export-related restrictions. Opus 5 now gives Anthropic another lever in a market where speed, pricing and deployment economics are shaping the fight for enterprise contracts as much as model quality is.