Opus 5 cuts frontier pricing, but completed tasks will settle the bill

Anthropic is positioning Claude Opus 5 as the default model for Claude Max and the strongest option in Claude Pro. The company says its capabilities approach Fable 5 at half the price. API pricing is $5 per million input tokens and $25 per million output tokens.

Opus 5 is moving from the showcase into daily work

Placement in the product line matters as much as specifications. Fable 5 remains the choice for the most demanding work and long running agent projects, while Opus 5 is meant to handle a wider share of activity in Max and Pro.

Anthropic is pushing high end capability into everyday coding, analysis and knowledge work. A model priced at the same level as Opus 4.8 may attract teams that previously reserved the more expensive option for selected tasks.

Budgets are measured by outcomes rather than individual tokens

The Verge reports that Opus 5 is slightly cheaper than OpenAI GPT-5.6. A Fast mode in research preview promises higher speed at twice the base price. The price table alone still cannot determine deployment economics.

If Opus 5 completes more steps without switching to Fable 5, repeating prompts or requiring manual repair, it can reduce the cost of a full task. If it produces longer outputs and needs more review, cheaper capability will remain expensive in operation.

Safety claims need evidence from real deployments

Anthropic calls Opus 5 its most aligned Opus model, with stronger cyber safeguards than Opus 4.8 and lower susceptibility to manipulation for misuse. It also points to independent testing with government partners.

For enterprise customers, those claims begin the evaluation rather than end it. Long agentic tasks require precise permissions, audit trails and escalation rules because a more capable model can perform more useful actions as well as more damaging mistakes.

Operations will reveal repairs and completed tasks

The most useful metrics will be the share of tasks completed without human intervention, total token consumption and review time. It will also matter whether Max and Pro users stop escalating work to Fable 5. Those numbers will show whether half price intelligence means cheaper work or merely a cheaper unit on a longer usage report.

Lilith's verdict

Anthropic is sending Opus 5 onto the regular shift with a promise of Fable 5 class performance. At the end of the week, a team lead will count finished tasks, manual repairs and review time rather than the discount per million tokens.