Anthropic releases Claude Opus 5 with cost-control toggle for enterprise users

Anthropic released Claude Opus 5 today as a new model aimed at everyday business needs. The release is notable for introducing a feature that allows users to toggle the effort level—low, medium, or high—that the model expends on completing a task or answering a prompt, enabling a balance between cost and capability. The model is also described as easier to use than its predecessors, requiring less back-and-forth interaction, and it can verify its own work and recover from errors without user intervention. This is Anthropic's fourth model release in less than two months, following Mythos 5, Fable 5, and Sonnet 5 in June.

The timing reflects growing pressure from enterprise customers over AI costs. Fable 5, the company's most powerful model, encountered two major obstacles: a high token burn rate that caused users to exceed budgets or incur large bills, and overly strict safety guardrails that prevented legitimate work in coding, cybersecurity, biological research, and AI model development. In June, the U.S. government temporarily imposed export controls on Fable 5 after Amazon researchers reported they could bypass its safeguards; Anthropic removed the model on June 12 and re-released it on June 30 after strengthening its security.

Opus 5 addresses these concerns directly. According to Anthropic, it "comes close" to Fable 5's capabilities but at half the price. For more advanced projects requiring autonomous operation over days, Anthropic still recommends Fable 5. On safety, Opus 5 adopts Fable 5's fallback mechanism: when the model declines a request due to safety concerns, the API automatically invokes another model, returning an answer rather than an error. Anthropic said Opus 5's safeguards are similar to Opus 4.8's in most respects, except in cybersecurity, where guardrails are stronger. Despite not being trained specifically for cybersecurity, the model "naturally improved in this area as a result of becoming more generally capable." Anthropic also described Opus 5 as "the most aligned Opus model" and "the least susceptible to being tricked into misuse."

Beyond cost and safety, Anthropic is positioning Opus 5 for the scientific research community, calling it its "most capable generally available model for scientific research" with particular strength in biology tasks. The release announcement also includes endorsements from law firms and software engineering teams, reaffirming these as core customer segments. The timing also comes as OpenAI marketing GPT-5.6, released on July 9, emphasized economical token usage.

Anthropic's release of Claude Opus 5 comes at a critical moment for the AI industry: enterprise customers are grappling with unexpectedly high AI usage costs, and questions about safety guardrails have created tension between protecting users and enabling legitimate work. Fable 5, Anthropic's previous flagship model, demonstrated both extremes—it was powerful enough that the U.S. government initially imposed export controls, but its high token burn rate and strict safety guardrails frustrated users in coding, cybersecurity, and biological research. Opus 5 targets the center of this friction: it offers near-Fable 5 capabilities at half the cost, addresses token bloat through the toggleable effort feature, and rebalances safety guardrails to allow more legitimate security work (notably, cybersecurity) while maintaining alignment protections. The automatic fallback mechanism also reflects lessons from Fable 5, ensuring users do not face failures when requests are declined. That Opus 5 is positioned as Anthropic's most capable model for scientific research suggests the company is also competing for academic and R&D customers beyond the traditional software engineering base.

No comments yet. Be the first to share your thoughts!