Anthropic released Claude Opus 5 on July 24, 2026. If you already pay for Claude, the practical news is short: it is now the default model on Claude Max and the strongest model available on Claude Pro, and the subscription price did not change. API pricing also held at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. You are getting a better model inside the plan you already have.
The more interesting part is a feature rather than a benchmark. Opus 5 introduces an effort setting that lets you decide how hard the model works on a given request — optimising for intelligence at one end, or conserving tokens for faster and cheaper results at the other. That is a genuinely new consumer-facing idea, and it is worth understanding because it changes what 'which model should I use' even means.
Quick summary: Claude Opus 5 launched July 24, 2026 at $5 per million input tokens and $25 per million output tokens — unchanged from Opus 4.8. It is the new default on Claude Max and the strongest model on Claude Pro, with no subscription price increase. A new effort setting (low, medium, high, xhigh, max) trades capability against token cost. A separate Fast mode runs around 2.5 times default speed at twice the base price on the Claude Platform. Anthropic's own benchmarks claim Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1 at lower cost per task, and lands within 0.5 percent of Fable 5 on CursorBench 3.2 at half the cost per task. It remains behind Mythos 5 on cybersecurity exploitation and biology research.
The Effort Setting Is the Real Story
Until now, choosing how much compute your question got meant choosing a different model. A quick factual lookup and a hard architectural problem both went to whichever model you had selected, and the only lever was switching models entirely. The effort setting collapses that decision into a dial on a single model: low, medium, high, xhigh and max. Low effort answers fast and cheap; max effort spends far more tokens thinking before it replies.
This matters more than it sounds. Most people either overpay by sending trivial questions to the most expensive model, or underpay by sending genuinely hard problems to a small one and getting a confident wrong answer. A single model with an effort dial means the right response to 'this is a hard question' is to turn the dial up rather than to remember which model was best. For anyone paying per token, it is also the most direct cost control Anthropic has shipped.
| Detail | Claude Opus 5 | What it means for you |
|---|---|---|
| Launch date | July 24, 2026 | Available now on Claude Max and Pro |
| API input price | $5 per million tokens | Unchanged from Opus 4.8 |
| API output price | $25 per million tokens | Unchanged from Opus 4.8 |
| Claude Max | New default model | You get it automatically, no price change |
| Claude Pro | Strongest model available | Better model inside your existing $20 plan |
| Effort setting | Low, medium, high, xhigh, max | Dial capability against token cost per request |
| Fast mode | About 2.5x speed | Costs twice the base price on the Claude Platform |
The Benchmarks, and How to Read Them
Every number below comes from Anthropic's own announcement. No independent reproduction exists yet, which is normal on launch day and worth stating plainly rather than presenting vendor claims as settled fact. With that caveat, the claims are: Opus 5 more than doubles Opus 4.8's performance on Frontier-Bench v0.1 at a lower cost per task; it scores within 0.5 percent of Fable 5's peak on CursorBench 3.2 at half the cost per task; it scores three times the next-best model on ARC-AGI 3; its Zapier AutomationBench pass rate is around 1.5 times the next-best model at the same cost; and it surpasses Fable 5's best OSWorld 2.0 result at just over a third of the cost. Source: Anthropic.
Notice the shape of those claims. Almost every one is framed as performance per unit cost rather than raw performance. Anthropic is not claiming Opus 5 is the most capable model in existence — it explicitly says Opus 5 remains behind Mythos 5 on cybersecurity exploitation tasks and biology research. The claim is that it gets close to frontier quality at a substantially lower cost per task. That is a positioning statement about value, and it is the honest way to read the launch.
What Actually Changes for You
- On Claude Max: nothing to do. Opus 5 is the new default, so you are already using it.
- On Claude Pro at $20 a month: it is now the strongest model you can select, at no price increase. This is the clearest win of the launch.
- On the free tier: no change from this announcement. Opus-class models remain a paid feature.
- Building on the API: prices held at $5 and $25 per million tokens, so any quality gain is a straight improvement in value. The effort setting is your new cost lever.
- If speed matters more than money: Fast mode runs about 2.5 times quicker at twice the base price on the Claude Platform.
- If you were on Opus 4.8 for cost reasons: the pricing is identical, so there is no reason to stay on the older model.
The Wider Context
This landed in a genuinely busy week. Google disclosed 950 million monthly Gemini users on July 22 and shipped three new Gemini models on July 21. Anthropic upgraded Claude's voice mode on July 23, and OpenAI brought full voice to the ChatGPT desktop app the same week. The competitive pattern across all of it is consistent: the frontier labs are competing on cost per unit of capability and on where the model can reach, rather than on headline intelligence alone.
For anyone actually choosing what to use, that convergence is the useful signal. When three vendors ship comparable quality within days of each other, the deciding factors become price, integration and which one handles your particular kind of work best — and those you can only learn by testing on your own tasks. That is the reason LumiChats keeps many current models under one login at a pay-per-day price: launch-day benchmarks tell you what a lab wants you to believe, and running the same real problem through several models tells you what is actually true for you.
01Do I have to pay more to use Claude Opus 5?
No. It is the new default model on Claude Max and the strongest model on Claude Pro, with no subscription price change. API pricing also held at $5 per million input tokens and $25 per million output tokens, identical to Opus 4.8.
02What does the effort setting actually do?
It lets you choose how much the model works on a request, across low, medium, high, xhigh and max. Higher effort spends more tokens reasoning before answering; lower effort is faster and cheaper. It is a way to match cost to how hard the question genuinely is.
03Is Opus 5 the most capable AI model available?
Not on everything, and Anthropic does not claim that. It says Opus 5 remains behind Mythos 5 on cybersecurity exploitation tasks and biology research. Its claims are mostly about performance per unit of cost — for example scoring within 0.5 percent of Fable 5 on CursorBench 3.2 at half the cost per task.
04What is Fast mode and is it worth it?
Fast mode runs at around 2.5 times the default speed and costs twice Opus 5's base price on the Claude Platform. It is worth it when latency matters more than token cost — interactive tools, live workflows — and not otherwise.
05Can free Claude users access Opus 5?
No. This announcement changes the model available on the Max and Pro paid tiers. Opus-class models remain a paid feature.
06Should I trust the launch benchmarks?
Treat them as vendor claims until independently reproduced. Every figure published on launch day comes from Anthropic's own testing. That is normal for the industry, but it is not the same as third-party verification, and independent results sometimes differ.
The bottom line: if you already pay for Claude, this is a free upgrade and you should simply use it. If you build on the API, the price held while the capability rose, and the new effort setting gives you a cost lever that did not exist before. And if you are comparing assistants, the honest read of Opus 5 is not that it beats everything — Anthropic says plainly that it does not — but that near-frontier quality keeps getting cheaper, which is the trend that matters most to anyone paying the bill.
