Anthropic has released Opus 5, the latest update to its series of large language models, which is designed to offer improved token efficiency rather than a significant leap in capabilities. The model is positioned as a cost-effective alternative to its more capable but expensive counterpart, Fable. According to Anthropic, Opus 5 performs at about the same level or slightly ahead of Fable in coding tasks, while being approximately half the cost. The company claims it is relatively good at finding cybersecurity vulnerabilities, but it is substantially behind Mythos 5 in exploiting those vulnerabilities due to its training decisions.
Benchmark results show that Opus 5 outperforms Opus 4.8 and OpenAI’s GPT-5.6-Sol in most tasks, with performance improvements appearing as iterative rather than radical. The model’s pricing is $5 per million input tokens and $25 per million output tokens, on par with its predecessor but cheaper than Fable. However, competition is intensifying, with the Chinese open-weight model Kimi K3 offering similar performance at $15 per million output tokens. Anthropic emphasized that Opus 5 lacks some of the controversial protections that Fable had, such as the 30-day data retention policy for review in case of an incident.
The release of Opus 5 comes amid growing interest in cost-effective alternatives to high-end models, with companies like Cursor and Meta developing model routers that automatically select the most appropriate model based on the task. This trend reflects a broader shift in the industry toward balancing performance with cost efficiency. Anthropic is expected to continue innovating in token efficiency to maintain its competitive edge in the evolving landscape of AI models.
Source: arstechnica
