Anthropic launched Opus 5 on July 24, positioning the new model as a cost-effective option close to its top-performing Fable 5 in key tasks. The release emphasizes coding improvements, autonomy, and reduced safeguards compared to prior flagship models.
Anthropic rolled out Opus 5 as an update to its Claude lineup. The model delivers performance near Fable 5 levels at half the price, with gains in coding, knowledge work, and scientific research such as organic chemistry and protein sequence analysis.
Users can adjust effort levels via a dial to balance thoroughness and speed. A research preview of fast mode offers 2.5 times faster output at twice the price of Opus 4. A fallback system switches to another model if safety filters block a prompt.
Pricing stays at $5 per million input tokens and $25 per million output tokens. Opus 5 avoids the 30-day data retention policy applied to Fable and was not trained on advanced cybersecurity exploitation tasks.
Anthropic says the model shows the lowest rates of deceptive behavior among its releases and intervenes on restricted prompts 85 percent less often than Fable 5.
