Anthropic has introduced Claude Opus 5, a new large language model that the company says delivers performance close to its flagship Fable 5 model while costing about half as much.
According to Anthropic, Opus 5 is its strongest model yet for knowledge-based work. The company also says it significantly improves complex coding capabilities compared to Opus 4.8 and performs better in scientific research tasks, including predicting how changes in protein sequences may affect molecular function.
Anthropic said it has strengthened Opus 5 against attempts to manipulate or trick the model. The company also claims the model shows its lowest levels of deceptive behavior to date.
Unlike Mythos, which is currently available only to selected organizations through Anthropic’s Project Glasswing initiative, Opus 5 was not specifically trained on cybersecurity-related tasks.
However, Anthropic said the model is naturally more capable than its predecessor in cybersecurity because of its broader improvements. This makes it more useful for identifying security vulnerabilities, although the company said it remains “substantially behind” its flagship model when it comes to exploiting those vulnerabilities.
Anthropic expects Opus 5 to become the primary model for everyday work.
The company acknowledged that some users found Fable 5 difficult to use because Claude automatically routed prompts on certain topics to the older Opus 4.8 model due to safety restrictions.
According to Anthropic, Opus 5’s safeguards are designed to support legitimate uses in cybersecurity and biology while maintaining targeted protections for a limited set of cyber-related tasks.
The company said its internal testing indicates that safety classifiers will intervene around 85% less often than they do with Fable 5.
Anthropic also confirmed that Opus 5 is not included in the company’s recently introduced 30-day data retention policy, which was announced alongside Fable 5 and Mythos.
Anthropic has kept API pricing for Opus 5 unchanged at $5 per one million input tokens and $25 per one million output tokens.
The company has also added a new “effort” setting that allows users to choose whether the model should prioritize thorough reasoning or faster, more token-efficient responses.
Access to Opus 5 is included with all of Anthropic’s paid subscription plans.
The launch comes as competition in the AI market continues to grow. Chinese AI company Moonshot AI recently introduced Kimi K3, which offers output pricing of $15 per one million output tokens and has been reported to deliver performance comparable to Fable 5 on some tasks. However, Anthropic noted that Opus 5 is available across all of its paid tiers, while Kimi’s subscription plans are not currently open for new sign-ups.
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.
Shares
