Anthropic Bets on Cheaper AI with New Claude Opus 5 Model

AI startup Anthropic has released Claude Opus 5, a new artificial intelligence model designed to offer high performance at a reduced cost. The company claims the model rivals its most powerful system, Claude Fable 5, but operates at half the price. This move directly addresses growing concerns among business customers about the significant expense associated with running advanced AI systems.

“

Opus 5 costs $5 per million input tokens and $25 per million output tokens, half the price of Fable 5, Anthropic’s most expensive publicly available offering.

Based in San Francisco, Anthropic positions Claude Opus 5 as a strong competitor to OpenAI's ChatGPT and Google's Gemini. Pricing for these AI models is typically based on usage, measured in tokens. Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, a substantial reduction from Fable 5's cost. Anthropic asserts that despite the lower price, Opus 5 achieves performance close to Fable 5 across a wide range of tasks.

“

The company said the new model performs close to Fable 5 across many tasks despite the lower cost.

The increasing scrutiny on AI companies' financial performance, especially following Anthropic and OpenAI's filings for stock market listings in June, has intensified the focus on cost-effectiveness. Wall Street is closely examining their business operations ahead of potential market debuts.

“

On cybersecurity, Anthropic said it deliberately avoided training Opus 5 on cyber tasks, and that the model remains behind Mythos 5, its most capable and most tightly restricted system, available only to select partners, on both offensive cyber work and biological research capabilities that could be misused.

Regarding cybersecurity, Anthropic stated that Opus 5 was intentionally not trained on cyber-specific tasks. It is positioned behind Mythos 5, Anthropic's most advanced and restricted system available to select partners, in both offensive cyber capabilities and biological research potential that could be misused. While Opus 5 demonstrates proficiency in identifying software flaws, comparable to Mythos 5, it significantly lags in its ability to weaponize these vulnerabilities. Anthropic reported that its automated safety filters intervene approximately 85% less frequently with Opus 5 than with Fable 5, citing the reduced risk profile.

“

The company said Opus 5 finds software flaws about as well as Mythos 5 but lags far behind in turning those flaws into something that could be weaponised into an attack.