Estimated read time: 7 minutes
Anthropic dropped Claude Opus 5 on July 24, and for once the headline isn’t just another benchmark chart nobody outside a lab will ever run. The part worth your attention as a small business owner is what Anthropic did to the price and the controls: roughly flagship-level intelligence at about half the cost of the last Opus, plus a dial that lets you decide how hard the model thinks before it opens its mouth. That combination quietly changes the math on when it’s worth pointing AI at real work instead of using it as a fancy autocomplete. Here’s what actually shipped, how it compares to OpenAI’s GPT-5.6, and whether any of it should change what you’re doing on Monday.
Table of Contents
What Anthropic Actually Launched
Claude Opus 5 is Anthropic’s new frontier workhorse, released July 24, 2026. The pitch in one sentence: it reaches close to Claude Fable 5-level intelligence — the top of Anthropic’s lineup — for roughly half the price of the previous Opus. It carries a 1 million-token context window (think a small filing cabinet of text in a single conversation), can produce up to 128,000 tokens in one go, and has a knowledge cutoff of May 2026, so it isn’t clueless about recent events.
On the money side, API pricing lands at $5 per million input tokens and $25 per million output tokens — the same sticker as Opus 4.8, except you’re getting materially more model for it. If you don’t touch the API and just pay for Claude like a normal human, Opus 5 is now the default on the Claude Max plan and the strongest option on Claude Pro. In other words, if you already subscribe, you’re probably using it right now whether or not you clicked a single button.
The Effort Toggle: The Feature That Touches Your Budget
The genuinely new lever is an effort setting with three positions: low, medium, and high. Opus 5 is also the first Opus that runs “adaptive thinking” by default, meaning it reasons through a problem before answering unless you tell it to keep things cheap and fast. That sounds like an engineering detail. It’s actually a budgeting feature.
Here’s why you care: that thinking time is billed at the output rate, and early users running the same task at high effort are reporting roughly double the output tokens of the previous Opus. So the toggle is a direct cost dial. Crank it to high for the proposal a client will actually read, and you get more careful, better-reasoned work at a higher price. Drop it to low for bulk, low-stakes chores — sorting inbound email, tagging support tickets, spitting out first-draft social captions — and you keep the bill sane. For a small business, that’s the entire game: matching what you spend to how much the answer actually matters. Most owners have been paying premium prices for tasks that never needed it.
How It Stacks Up Against GPT-5.6
The unavoidable comparison is OpenAI’s GPT-5.6 Sol. Early published benchmarks give Opus 5 the edge on agentic coding, computer use, and multi-step “go do this whole task” workflows — the category that matters if you want AI to operate software on your behalf rather than just answer questions. GPT-5.6 Sol still wins a couple of categories and remains a genuinely excellent model; this is not a blowout. On price the two are close, with Opus 5 at $5/$25 per million tokens versus roughly $5/$30 for GPT-5.6 Sol, so Anthropic is a touch cheaper on output.
But here’s the honest read for a small business: the benchmark leaderboard is a tie-breaker, not a decision. Both of these models are far smarter than the work most of us hand them. The questions that actually decide it are which tools you already pay for, which one your team will genuinely use without friction, and whether the thing plugs into the software you already run. We broke down the OpenAI side of this in our guide to the GPT-5.6 tiers, and the short version holds: pick the one that disappears into your workflow.
What This Actually Means for a Small Business
Strip away the launch-day noise and three practical shifts are left.
1. Agentic work gets more trustworthy. Higher computer-use and task-completion scores mean the “have the AI pull the report, fill the form, book the thing” use cases get less flaky. You still supervise — nobody should let a model loose on live systems unwatched — but the failure rate that made these features a party trick is dropping toward “actually useful.”
2. AI finally becomes budgetable. The effort toggle means you’re no longer paying one flat premium for every task. Run cheap-and-cheerful for volume, expensive-and-careful for anything a customer sees. That’s the difference between an AI bill you dread and one you can forecast.
3. Longer memory means fewer re-explanations. A million-token context window lets you hand the model an entire SOP library, a tangled three-month customer thread, or a quarter’s worth of invoices and have it keep the plot instead of forgetting the beginning by the time it reaches the end. For anyone who’s spent twenty minutes re-pasting context into a chatbot, that alone is worth the upgrade.
Should You Switch? A Quick Gut-Check
If you already pay for Claude Pro or Max, there’s nothing to switch — you’re on Opus 5 already. The move is to spend ten minutes re-running a task that used to disappoint you and see if the ceiling moved. If you’re standardized on ChatGPT and your team is happy, a rival’s launch day is not a reason to rip out your stack; switching costs — retraining people, rebuilding prompts, reconnecting tools — are real and rarely worth chasing a few benchmark points.
The smarter posture for most owners isn’t crowning a winner at all. It’s knowing which model to reach for per job, and building a light habit where AI output gets a human glance before it ships to a client. The tools keep leapfrogging each other every few weeks. Your process for using them well is the part that compounds. If you’re still mapping which AI does what, our rundown of the best AI writing tools for small business is a decent place to start.
Frequently Asked Questions
How much does Claude Opus 5 cost? If you use Claude as a consumer, it’s bundled into Claude Pro (around $20/month) and is the default model on Claude Max. Through the API, it’s $5 per million input tokens and $25 per million output tokens.
Is Claude Opus 5 better than GPT-5.6? On agentic-coding and computer-use benchmarks, early numbers favor Opus 5. GPT-5.6 Sol leads in a couple of categories and is still excellent. For everyday business tasks the two are close enough that workflow fit matters more than the scoreboard.
What is the effort toggle? A low/medium/high setting that controls how much the model “thinks” before answering. Higher effort means better answers, more tokens used, and a higher cost. Lower effort is cheaper and faster for simple work.
I pay for Claude already. Do I need to do anything? No. If you’re on Max or Pro, Opus 5 is already your default or top-tier option. Just try re-running a task or two to feel the difference.
Related Coverage
- GPT-5.6 Tiers Explained for Small Business — the OpenAI side of the same decision, tier by tier.
- AI Agents Explained: The 2026 Rundown — what “agentic” actually means before you trust one with real tasks.
- Best AI Writing Tools for Small Business — which model to reach for when you’re the one on deadline.
Faceted Media Magazine covers business, AI, and entrepreneurship for the people building what’s next.