The news that matters for a business
If you lead AI adoption in your company, the ARC-AGI 3 benchmark doesn't change your week. The effort dial does. It's the Opus 5 feature that puts a direct lever on cost and quality in your hands, task by task.
Until now the trade-off was: flagship model for quality, but expensive at volume; small model for cost, but with more errors. Opus 5 makes the choice continuous instead of binary. You decide how much power to spend on each request. Below we look at what that means in practice on three fronts: costs, agents, adoption. If you need the general picture first, start from what Opus 5 is.
The effort dial as a cost lever
The Opus 5 list price is the same as Opus 4.8: $5/$25 per million tokens. But the cost you see on the bill depends on how many tokens you consume, and here the effort dial works for you.
A concrete example. An email classification and reply flow: the classification is simple, drafting a reply to a complex complaint is delicate. With a single model at variable effort, you give low effort to the first part and high effort to the second. Before, you'd have had to orchestrate two different models, with more complexity and more breaking points. Fewer pieces, same quality, cost under control. To measure the impact on your volumes, cross-reference the pricing simulator with the guide on how much Claude costs for business.
What changes for your agents
If you have agents in production — document copilots, CRM automations, internal assistants — Opus 5 touches them in two ways. At high effort it closes tasks that used to fail: multi-step planning, using several tools in sequence, edge cases. Fewer hand-offs to humans, more real autonomy.
At low effort it saves you money on repetitive tasks without a perceptible drop in quality. And for anyone building on the API there are two useful betas: changing the agent's tools mid-conversation without losing the cache, and automatic fallbacks that avoid abrupt interruptions. If you're weighing where to put your first agent, our guide on how to implement AI agents with Claude is the starting point.
Where does Opus 5 shift the ROI in your processes?
30 minutes to discuss your specific case.
Adoption: why the right model isn't enough
Here an uncomfortable thing has to be said. A better model doesn't, on its own, produce more value. It produces it if people use it well on the right processes. Opus 5 lowers the barrier — more reliable, cheaper to run — but adoption still makes the difference.
In practice: you need to understand which tasks are worth automating, build the prompts and workflows on the team's real documents, and measure usage. It's exactly the work we do with our training and adoption programs. The new model is an opportunity to redo this mapping, not a substitute for it.
How to review the ROI of your flows
Every time a model comes out that shifts the cost/quality equation, the business cases you made before need redoing. Tasks that six months ago weren't worth automating — too expensive or too unreliable — with Opus 5 at tuned effort could become worth it.
The method is simple: take the flows where AI already runs, re-measure cost per task and success rate with Opus 5 at the various effort levels, and compare with the equivalent human cost. Then look at the flows you'd discarded: some might be back within threshold. For the how, we have a dedicated guide on how to measure AI ROI in your company.
Where to start
If you already use Claude, the first step is to review model routing in light of the effort dial and re-run your evals on the flows in production. If you're starting now, the point isn't "Opus 5 or not": it's to identify the two or three processes where AI produces measurable value, and build there.
At Maverick AI we do both: we pick the right model and effort for each task and bring AI agents into production without burning budget, with the training the team needs to actually use them. If you want to understand where Opus 5 shifts the ROI in your processes, let's talk.