What makes Opus 5 achieve rival performance at half the cost?

Question

Accepted Answer

Anthropic says Claude Opus 5 matches top performance at half the cost mainly through lower computing-resource use and cheaper token pricing, not by a disclosed architecture change

Anthropic released Claude Opus 5 on July 25 and positioned it as approaching the performance of its most powerful system at half the price. The company attributes the cost difference to Opus 5 consuming fewer computing resources than earlier versions, while also setting a lower price of US$5 per million input tokens and US$25 per million output tokens. It also said Opus 5 performs close to its Fable 5 model across many tasks despite the lower cost.
Opus 5’s cost is framed around computing resources and token charges
Anthropic said Opus 5 consumes fewer computing resources than earlier versions, which is presented as the basis for its improved cost profile. In practical terms for business customers, pricing is tied to tokens, with Opus 5 charged at US$5 per million input tokens and US$25 per million output tokens. The company also stated this is half the price of Fable 5, described as its most expensive publicly available offering.
The company links performance to Fable 5 despite the lower price
Anthropic said Opus 5 tops industry benchmarks for coding and knowledge work while remaining close in performance to Fable 5 across many tasks. This is offered as the reason customers can run advanced work at a lower unit cost without losing comparable capability. The comparison is explicit: Opus 5 is priced at half the cost of Fable 5 while aiming to deliver similar results across a range of tasks.
Cyber capability limits are part of how the model is positioned, but not the cost mechanism
On cybersecurity, Anthropic said it deliberately avoided training Opus 5 on cyber tasks. It also said Opus 5 remains behind Mythos 5, Anthropic’s most capable and most tightly restricted system, on offensive cyber work and biological research capabilities that could be misused. Anthropic further said Opus 5 finds software flaws about as well as Mythos 5, but lags far behind in turning flaws into something that could be weaponized into an attack.
Safety interventions differ versus Fable 5, but technical changes enabling the cost ratio were not detailed
Anthropic said its automated safety filters will intervene roughly 85 percent less often on Opus 5 than on Fable 5. However, the available information does not spell out what specific training methodologies or model architecture changes produce the “fewer computing resources” result. The cost-performance claim is therefore supported by the company’s stated resource reduction and the published token prices, rather than by disclosed technical innovations.

Reading history

Hide →

This history is cleared when the page is closed.

What makes Opus 5 achieve rival performance at half the cost?

Loading

Was this article helpful?

Helpful

Not helpful

Report an issue

1

0

0

Share with friends