Anthropic launched Claude Opus 5 today, and the headline number is simple: you get close to Fable 5 quality at half the price. Fable 5, Anthropic's most powerful publicly available model, costs $10 per million input tokens and $50 per million output tokens. Opus 5 comes in at $5 and $25, the same price as its predecessor, Opus 4.8.
The price staying flat while capability jumped substantially is the real story here. On benchmark tests for coding and complex knowledge tasks, Opus 5 not only matched Fable 5, it beat it on several. On one coding benchmark, Opus 5 scored 43.3% compared to Fable 5's 33.7%. On a test measuring computer-use tasks, it cleared Fable 5's best score at roughly a third of the cost.
This matters to any business operator building or buying AI tools, because the unit economics of AI change dramatically when a more capable model costs the same as what you were already paying. Legal firm Harvey, for example, reported that Opus 5 matched the quality of the previous model running at full power while cutting average usage costs by 26%. That is not a lab result, that is a live production number.
Anthropomorphic also added an effort dial. Users can set it to low, medium, or high depending on how thorough they need the model to be for a given task. Simpler tasks can run cheap and fast; complex ones get more processing power. This kind of cost control has been missing from most AI tools, and it gives finance and operations teams something they can actually manage.
To understand why Opus 5 exists the way it does, you need to know what happened with Fable 5 in June. Fable 5, Anthropic's most advanced public model, was taken offline by US government order just three days after launch over fears that researchers had found a way to trick it into helping locate software vulnerabilities. The government's export control directive forced Anthropic to disable both Fable 5 and its even more powerful government-only model, Mythos 5, for all users globally. After weeks of negotiations, both models came back, but with significantly tighter restrictions.
Those restrictions frustrated many legitimate users. Fable 5's safety filters were triggering on ordinary professional tasks, blocking work that had nothing to do with hacking or security threats. Anthropic built Opus 5 with a deliberately lighter set of restrictions, designed to stay well below the capability level that caught the government's attention. Anthropic says it ran Opus 5 by its government partners before launch.
The result is a model that Anthropic expects will be the one most users actually reach for daily. Fable 5 remains the better choice for the most complex, long-running tasks and AI agent projects. But for everything from reading contracts and drafting reports to writing and debugging code, Opus 5 is now the practical default.
Anthropomorphic is also offering a Fast mode, running at 2.5 times the default speed for double the price. That is still cheaper than Fable 5 at standard speed, so for teams that need quick turnaround on large volumes of work, the math can still work in their favor.
The competitive backdrop is real. Chinese AI company Moonshot recently released its Kimi K3 model, which reportedly delivers Fable-level performance on some tasks at $15 per million output tokens. That is cheaper than Fable 5 but more expensive than Opus 5, and Kimi does not yet offer paid subscription plans outside China. The pressure is visible in how Anthropic framed this launch: not around raw capability, but around value per dollar. That shift in framing reflects where enterprise buyers are right now.