Anthropic Launches Opus 5: A Balance of Cost and Performance

Anthropic has officially released its newest artificial intelligence model, Opus 5. This update is designed to offer a balance between high-level performance and lower costs for business customers. While the company describes the model as a significant step forward, it is not a major breakthrough in technology. Instead, the main goal of Opus 5 is to provide a tool that performs nearly as well as the company’s flagship model, Fable 5, but at approximately half the price.

According to The Verge, the release of Opus 5 comes shortly after a period of difficult discussions between Anthropic and the US government. These talks were focused on cybersecurity—the practice of protecting computer systems from digital attacks. The government had previously raised concerns about the risks associated with powerful AI models, leading to the temporary removal of the Fable 5 and Mythos 5 models from public use. After these models were updated with stronger safety rules, they were brought back online. Anthropic spokesperson Danielle Ghiglieri stated that the company continues to work with government partners to conduct independent testing on all new models, including Opus 5.

In terms of technical capabilities, Ars Technica reports that Opus 5 performs slightly better than its predecessor, Opus 4.8, and competes well against OpenAI’s GPT-5.6-Sol. The model is specifically marketed toward enterprise customers—large businesses that need AI for complex tasks like knowledge management and biology research. While Fable 5 remains the best choice for the most ambitious and long-term AI projects, Anthropic claims that Opus 5 is the most "aligned" model it has ever created. This means it is designed to follow human instructions safely and is the least likely to be tricked into misuse.

However, there are clear differences in how the models are built. Ars Technica noted that Anthropic intentionally chose not to include the most advanced cybersecurity training in Opus 5. As a result, the model is not as strong as Mythos 5 when it comes to finding or exploiting digital vulnerabilities. Additionally, Opus 5 does not include some of the controversial data-tracking policies found in Fable 5, such as the rule that keeps user data for 30 days to review in case of a security incident.

For customers, the pricing structure remains a key part of the announcement. The Verge reports that Opus 5 costs $5 per million tokens for input and $25 per million tokens for output. A token is a basic unit of text that AI models process. This pricing is the same as the previous version, Opus 4.8, but is much cheaper than Fable 5. The company also introduced a new "Fast mode" for users who need higher speeds, though this feature costs double the standard price. Furthermore, users can now choose to have the system automatically switch to a simpler, lower-tier model if the security rules prevent Opus 5 from answering a specific request.

This release also seems to address recent industry debates regarding the high costs of running AI systems. There have been concerns that AI labs were passing their high operational expenses onto customers in hidden ways. By keeping the price steady while improving efficiency, Anthropic is trying to show that it can provide powerful tools without making them unaffordable for businesses. As the AI industry continues to grow, the relationship between developers like Anthropic and government regulators remains a central part of the conversation. With Opus 5, the company is attempting to prove that it can innovate while maintaining the safety standards required by the government.