Anthropic Says Opus 5 Nears Fable at Half the Price
New model approaches Fable 5 on some coding tasks while charging half its token rates.
Topics
Anthropic has released Claude Opus 5, saying the artificial intelligence model approaches its more powerful Fable 5 system on some coding and knowledge-work tasks at half the list price.
Opus 5 became available on Friday, 24 July, at $5 per million input tokens and $25 per million output tokens. Anthropic charges $5 for every million tokens users send to the model and $25 for every million tokens it generates in response. A token is usually a word, part of a word or a punctuation mark, so the final cost depends on how much text the model processes and produces.
The rates are unchanged from its predecessor, Opus 4.8, and half Fable 5’s rates of $10 and $50. Opus 5 is now the default model for Claude Max subscribers and the strongest model offered under the Claude Pro plan.
Anthropic also claimed lower costs for completing individual benchmark tasks. On CursorBench 3.2, Opus 5 at its maximum effort setting scored within 0.5% of Fable 5’s best result while costing about half as much per task.
On Frontier-Bench v0.1, a software-engineering test, it more than doubled Opus 4.8’s performance while completing tasks at a lower cost. Anthropic said Opus 5 recorded the strongest result among the models it tested.
The findings are based largely on Anthropic’s own evaluations and early customer testing and have not been independently verified. In some assessments, requests blocked by Opus 5’s safety systems were redirected to Opus 4.8, which could affect direct comparisons.
Anthropic said the model was designed to check its work more carefully, correct errors and continue working through complex tasks without frequent human intervention.
In one evaluation, Opus 5 was asked to recreate a machine component from a drawing but was not given a direct way to inspect the image. The model built its own computer-vision pipeline, extracted the geometry and produced a three-dimensional model, the company said.
It also performed strongly on business automation and computer-use tests. On Zapier AutomationBench, Opus 5 completed about 1.5 times as many tasks as the next-best model at a similar cost, while on OSWorld 2.0 it surpassed Fable 5’s best result at just over one-third of the cost, according to Anthropic.
The model showed gains in scientific research, including structural biology, chemistry and bioinformatics. Anthropic said Opus 5 improved on Opus 4.8 by 10.2 percentage points on an internal organic chemistry test and by 7.7 points on a protein-function task.
Opus 5 remains behind Anthropic’s Mythos 5 model in offensive cybersecurity and advanced biology work. The company said the two models were similarly capable of identifying software vulnerabilities, but Opus 5 was substantially less successful at turning those weaknesses into working exploits.
Anthropic said Opus 5 can analyze source code for security flaws but blocks penetration testing, binary scanning and exploit generation for most users. Approved organizations in its Cyber Verification Program can access a version with fewer restrictions.
