Claude Opus 5 is here: what it means for your firm

Anthropic released Claude Opus 5 on Friday 24 July, and for Australian professional services firms the headline is simple: the strongest Claude model most firms will ever need is now the model that comes with the subscriptions many of them already pay for. It is built to check its own work, it costs about half what Anthropic's top-end model does, and it comes with a new control that lets you trade cost against capability on each task. Here is what changed, and what it is worth doing about it this month.

What did Anthropic actually release?

Claude Opus 5 is the newest version of Anthropic's Opus line, the model family that sits behind the Claude chat product most professionals use day to day. TechCrunch reports that it landed just two months after its predecessor, which tells you something about the pace the major AI labs are now setting. Blink and the tool your firm standardised on has been replaced by something meaningfully better.

According to SiliconANGLE's coverage, Opus 5 is now the default model on Claude's Max tier, is available on the Pro plan, and can be reached through the API for firms that build their own tools. In other words, if your firm already subscribes to Claude, you are not waiting for a rollout. It is there now.

The performance claims are notable even by AI launch standards. SiliconANGLE reports that Opus 5 outperformed Anthropic's larger flagship model on 8 of 13 benchmarks, including a knowledge-work benchmark, which is the category closest to what professional services firms actually do all day: reading, drafting, analysing and checking.

The feature that matters: it checks its own work

Every AI vendor claims their new model is smarter. The more useful claim here, repeated across Anthropic's materials and the independent coverage, is that Opus 5 is better at verifying its own output and iterating until it gets things right. Fortune reports that the model "requires less back and forth" than its predecessors, verifies its own work and recovers from errors on its own, and that early adopters highlighted in the launch include law firms.

For a partner or practice manager, this addresses the single biggest complaint we hear in training sessions: that AI output is a promising first draft that still needs three rounds of correction. A model that catches more of its own mistakes before you see them shifts the economics of using it. To be clear, it does not shift your professional obligations. A lawyer, accountant or architect still owns every document that goes out the door, and review remains non-negotiable. But reviewing work that is mostly right is a very different job from repairing work that is half right.

You can now dial effort up and down

Opus 5 introduces an effort setting with three levels: low, medium and high. Fortune describes it as a way to balance computational cost against response quality on a per-task basis.

This sounds like a technical detail. It is actually a workflow decision, and firms should treat it that way. Routine tasks (summarising a rental statement, tidying up file notes, first-pass triage of a document bundle) do not need maximum effort, and running them on a lower setting is faster and cheaper. Complex tasks (analysing an unusual contract clause, stress-testing a financial model, working through a complicated compliance question) justify the high setting. The firms that get the most out of this will be the ones that give staff simple guidance on which setting suits which kind of work, rather than leaving everyone to guess.

What does it cost?

Both Fortune and SiliconANGLE report that Opus 5 delivers capability close to Anthropic's top-tier model at roughly half the price. For subscribers the practical point is simpler: if you are on Claude Pro or Max, Opus 5 is included in what you already pay, and on Max it is now the default. Firms building on the API get the pricing benefit directly, which matters if you have internal tools or a technology partner running Claude behind the scenes.

There is also a safety story worth noting for risk and compliance conversations. Anthropic describes Opus 5 as its safest model to date, with less deceptive behaviour in testing and guardrails that trigger far less often on legitimate work. SiliconANGLE reports the safety classifiers are expected to engage about 85 per cent less often than on Anthropic's frontier model. For firms, fewer false refusals on ordinary professional tasks means fewer frustrated staff quietly switching to unapproved tools, which is exactly the shadow-AI problem most firms are trying to stamp out.

Why this matters for your firm

Three practical implications stand out.

• The capability floor just rose for everyone. The model doing the best knowledge work in independent benchmarks is now the default on a consumer-grade subscription. Your competitors have access to exactly the same thing. The differentiator is no longer access, it is how well your people use it.
• The refresh cycle is now measured in weeks. Four models in under two months from one vendor means "we evaluated AI last year" is no longer a current position. Firms need a lightweight way to re-test their standard tasks when a major model lands, not an annual review.
• Cost control is becoming a skill. Effort settings, model choices and plan tiers all now affect what AI costs your practice. Someone in the firm should own that question, the same way someone owns your practice management software licensing.

How to start this week

If your firm already uses Claude, open the model picker and confirm you are on Opus 5, then re-run two or three of your standard tasks (a letter, a summary, a review checklist) and compare the output against what you were getting last month. Note where the new model needs less correction, and update your prompt library or internal guidance to match. If your firm is still deciding on tools, this launch is a good moment to run a structured trial, because the gap between "we dabble" and "we have a standard way of working" is where most of the productivity benefit sits.

And whichever camp you are in, make sure someone owns the watching brief. The pace of these releases is not slowing down.

Want help making sense of releases like this?

Keeping up is a job in itself, and most partners do not have the spare hours. Our live sessions walk through what each major release actually changes for professional practices, with demonstrations on realistic firm workflows rather than vendor slideware. See what is coming up on our upcoming webinars page, or if you want this sort of guidance on tap for your whole firm, our consulting and in-house training works directly with practices across law, accounting, architecture, real estate and healthcare.

Sources: TechCrunch, SiliconANGLE, Fortune.