Claude Opus 5 Is Here: What's New, What It Costs, and Why It Matters

Anthropic released Claude Opus 5 on July 24, 2026, and the short version is this: it gets close to the intelligence of the top-tier Fable 5 model at half the price, and it costs the same as the Opus model it replaces, $5 per million input tokens and $25 per million output tokens. For anyone building or running software with AI, that means frontier-level coding and agent work at a price you can actually put into production. Here is what changed, what it costs, and where it fits.

Key takeaways

• Opus 5 launched on July 24, 2026, priced at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8.
• It approaches Fable 5 intelligence at roughly half the cost, and it is now the default model on Claude Max and the strongest model on Claude Pro.
• On Anthropic's coding and agent evaluations, Frontier-Bench, CursorBench, OSWorld, and Zapier AutomationBench, it leads, but it still trails Mythos 5 on cybersecurity tasks.
• The real story is efficiency and judgment: it finishes harder tasks in fewer tokens and turns, and it checks its own work before handing it back.
• For SaaS and AI teams, the practical win is cheaper agentic automation and coding that is finally reliable enough to ship.

What is Claude Opus 5?

Opus is Anthropic's high-capability, everyday tier, sitting below the frontier Fable and Mythos models but built to be run all day without wrecking your budget. Opus 5 is the newest version, and it is available across Claude.ai, the Claude API as claude-opus-5, Claude Code, and Claude Cowork. It is the new default on Claude Max and the strongest option on Claude Pro.

The pitch is not "the smartest model on earth." It is "almost the smartest, at a price and speed you can use in real products." There is also a Fast mode that runs about 2.5 times the default speed for twice the base price, which is useful when latency matters more than cost.

What is actually new in Opus 5?

The headline is that Opus 5 delivers much better performance for the same price as its predecessor. A few concrete points from Anthropic's own evaluations:

• On Frontier-Bench v0.1, a software engineering benchmark, Opus 5 surpasses every other model and more than doubles Opus 4.8's score at a lower cost per task.
• On CursorBench 3.2 at maximum effort, it lands within 0.5% of Fable 5's peak score while costing about half as much per task.
• On ARC-AGI 3, which tests novel problem solving, its score is roughly three times the next-best model.
• On Zapier AutomationBench, which checks whether a model can finish a business task end to end, its pass rate is around 1.5 times the next-best model for the same cost.
• On OSWorld 2.0, a computer-use benchmark, it beats every model at a given cost and passes Fable 5's best result at just over a third of the cost.

Numbers aside, the change people notice is judgment. Opus 5 verifies its work and iterates instead of declaring victory early. In Anthropic's testing it wrote its own computer-vision pipeline to read a drawing it was not allowed to view, found the root cause of an open-source bug that a community patch had missed, and built a market-data feed in a single session, complete with its own test harness. That kind of "finish the job" behavior is what makes an agent worth trusting.

It is also Anthropic's most aligned model to date, with the lowest measured rate of deceptive behavior in its recent lineup. For teams putting AI in front of customers, that reliability is not a nice-to-have.

How much does Claude Opus 5 cost?

Pricing is $5 per million input tokens and $25 per million output tokens, identical to Opus 4.8. Fast mode is available at twice that base price for roughly 2.5 times the speed. There is also an effort setting you can turn down to conserve tokens when a task does not need maximum reasoning, which is the easiest lever most teams miss.

The important cost point is not the sticker price, it is the total. Because Opus 5 tends to use fewer turns and fewer tokens to finish the same work, the real cost per completed task often comes in below models that look cheaper per token. When you budget for AI, measure cost per finished task, not cost per million tokens.

Why does Opus 5 matter for SaaS and AI teams?

If you run a software product, three things get easier. Agentic automation, the kind where a model reads data, decides, and acts across several steps, becomes cheaper and more dependable, which is exactly what the AutomationBench result points to. Coding assistance gets closer to senior-engineer quality, so a smaller team ships more. And customer-facing features get safer, thanks to the alignment gains.

I test tools like this constantly for the SaaS and AI product videos I make, and the pattern with Opus 5 is clear: it is the first Opus model I would comfortably hand a messy, multi-step task and expect a clean result. That is the difference between a demo that impresses and a feature that survives contact with real users.

Should you switch to Opus 5?

If you are already on Opus 4.8, this is a straightforward upgrade at the same price, so switch and test. If you use a cheaper model to save money, try Opus 5 at a lower effort setting before you assume it is out of budget, the cost-per-task math often favors it. The one caveat: if your work depends on frontier cybersecurity or advanced biology research, Mythos 5 still leads there. For nearly everything else a SaaS or AI team does day to day, Opus 5 is the new default worth reaching for.

Frequently asked questions

Is Claude Opus 5 better than Fable 5?

Not quite, and it is not trying to be. On several of Anthropic's coding and agent evaluations Opus 5 lands within a fraction of a percent of Fable 5, but at about half the cost. Fable 5 is still the frontier model; Opus 5 is the one most teams should actually run, because the price-to-performance ratio is far better for everyday work.

How much does Claude Opus 5 cost?

It is $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. Fast mode costs twice the base price for about 2.5 times the speed. Because it finishes tasks in fewer tokens and turns, the real cost per completed task is often lower than cheaper-looking models.

What is Claude Opus 5 best for?

Agentic coding, multi-step automation, and knowledge work where the model has to plan, act, and check itself. It is strong at software engineering, business automation, computer use, and scientific analysis. It is not the top pick for offensive cybersecurity or the hardest autonomous biology research, where Mythos 5 remains ahead.