Anthropic Launches Opus 5, Outperforming Fable 5 in Benchmarks
On Friday, Anthropic introduced its latest large language model, Opus 5, which outperforms the recently launched Fable 5 on several benchmarks. Despite being smaller than Fable 5, Opus is more cost-effective and less restrictive, making it a potentially preferred option for most applications. The release follows closely after Opus 4.8 in July and comes as part of Anthropic's broader model upgrades this year. Anthropic highlighted that Opus 5 excels at verifying its work and iterating until successful, with examples including the creation of a computer vision pipeline from an incomplete prompt. Unlike Fable 5, it is not subject to a 30-day data retention policy, addressing privacy concerns for some users. However, safeguards remain in place for cybersecurity tasks like exploit generation. To improve user experience when safety classifiers are triggered, Anthropic introduced Automatic Fallbacks—a beta feature that routes requests to less powerful models instead of returning errors.
Topics: AI Anthropic data privacy model launch Opus 5