Claude Opus 5 Explained: Which Plan Gets It and What It Costs

Anthropic shipped its fourth new model in under two months today, and this one is the first in a while that changes something for people who are already paying. On July 24, 2026 Anthropic released Claude Opus 5, calling it "the new default model on Claude Max, and the strongest model on Claude Pro." The consumer question underneath the benchmark charts is simple: do you get it, does it cost more, and does anything about your Claude actually feel different?

Key Takeaways

• Claude Opus 5 is available today, July 24, 2026, across Claude on the web, mobile, desktop, Claude Code and Cowork.
• The free plan does not get it. Opus is a paid-plan model, and always has been. Free Claude stays on Sonnet 5.
• On Pro ($20/month) it is included at no extra cost -- and it is now the best model your $20 covers.
• On Max (from $100/month) it is the new default, so you do not have to select it.
• It is not better than Fable 5. Anthropic says it comes "close to the frontier intelligence of Claude Fable 5 at half the price."
• The price to developers did not move: $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8.
• It knows more recent things. Opus 5's training data runs to May 2026, versus January 2026 for Sonnet 5 and Fable 5.
• You cannot turn its thinking off. Anthropic's help centre says extended thinking cannot be disabled in Claude when you are using Opus 5.

Do You Get It? Availability by Plan

Here is the honest picture as of July 24, 2026, read off Claude's pricing page and Anthropic's announcement.

The line that matters most is the Pro row, and it lands with unusual timing. Five days ago, on July 19, 2026, Anthropic's promotion that included Fable 5 inside Pro's usage limits expired, leaving Pro subscribers paying separately with usage credits if they wanted the top model. Today the best model that Pro does include got meaningfully better, at no extra charge. If you were annoyed on Monday, today is the compensation.

Is It Better Than the Claude You Were Using?

Yes, if you were on Opus 4.8, and Anthropic is unusually specific about how. On its own evaluations, Opus 5 "more than doubles Opus 4.8's performance" on the Frontier-Bench v0.1 coding benchmark at a lower cost per task, scores three times the next-best model on ARC-AGI 3 (a test of solving problems it has never seen), and roughly 1.5 times the next-best model's pass rate on Zapier's AutomationBench for completing business tasks end to end.

Treat all of those as vendor-reported numbers on tests Anthropic chose. The claim that is easier to sanity-check is the positioning one: Anthropic itself says Opus 5 remains behind its restricted Mythos 5 model on cybersecurity, and only comes "close to" Fable 5 overall. So this is a large step up in the tier most subscribers actually use, not a new crown.

For everyday work -- drafting, editing, reading long documents, thinking through a decision -- the difference you are most likely to notice is not raw intelligence. It is the three things below.

Three Things That Feel Different Day to Day

It knows about more recent events. Per Anthropic's help centre, Opus 5 was trained on data up to May 2026. Sonnet 5 and Fable 5 both stop at January 2026. Four months of extra background knowledge is the single most noticeable everyday upgrade, because it means fewer "as of my knowledge cutoff" detours on anything from this spring.

It always shows its thinking. Anthropic's settings guide states plainly that extended thinking cannot be turned off when you use Opus 5 in Claude. You can still change the effort level -- low, medium, high, extra high or max -- from the same menu next to the send button. Lower effort answers faster and stretches your usage limits further, which is the setting most people should actually touch.

It may swap models on you mid-chat. More on that next, because it is the one genuinely confusing behaviour in this release.

Why Claude Might Switch Models in the Middle of Your Conversation

Because a narrow set of security-related requests get re-run on an older model. Anthropic runs automatic safety checks on every Opus 5 request, and a help article published today spells out what happens when one trips: Claude visibly falls back from Opus 5 to Opus 4.8, tells you it switched, and labels the answer with the model that wrote it.

Three details worth knowing before it happens to you:

• The picker stays switched. After a fallback, the conversation stays on the less capable model until you manually switch back.
• It is not only your typing that gets checked. Anthropic says the checks review "everything the model reads" -- memory, connector content, web search results and files -- so a fallback can be triggered by something you did not write.
• You can turn it off. Settings > Capabilities, then toggle "Switch models when a message is flagged" off. With it off, a flagged request pauses the conversation instead of quietly changing model.

For normal use this should be rare: Anthropic says the triggers are things like exploit generation, penetration testing and binary vulnerability scanning, and that in early testing Opus 5 hit these fallbacks 85% less often than Fable 5. Scanning your own source code for bugs, triaging security issues and writing secure code are all still fine.

One related change that fixes a real annoyance: biology and chemistry questions blocked on Fable 5 now route to Opus 5 rather than Opus 4.8, so the fallback you land on is a much better model than it was yesterday.

What It Costs, and How That Compares

Nothing changed for subscribers, and nothing changed for developers either. Opus 5 lists at $5 per million input tokens and $25 per million output tokens -- identical to Opus 4.8. Anthropic also offers a Fast mode that runs about 2.5 times the default speed at twice the base price.

As a subscriber you never see token prices -- you pay $20 or $100 a month and hit usage limits instead. The reason the flat price matters to you is what it signals: Anthropic passed a capability jump through without raising anything, which is the opposite of what happened to Pro users with Fable 5 last week.

Worth noting for the privacy-minded: Anthropic says Opus 5, like previous Opus models, has no data retention requirements for general access, and every consumer plan is opt-out for model training.

Should You Care?

If you are on Claude Free: no, and nothing is being taken away. You keep Sonnet 5, which is still one of the better free models available. Today changes nothing on your plan. If you want the honest picture of what free tiers give you in 2026, we compare them in the best free AI chatbot.

If you are on Claude Pro: yes, and it is the reason to stay. Open the model picker and select Opus 5 for your harder work -- long documents, multi-step reasoning, anything you would previously have reached for Fable 5 to do. It is included, Fable 5 is not, and the gap between them is now small.

If you are on Claude Max: you already have it, whether you noticed or not. Opus 5 is the new default there. The main thing to know is that you can drop the effort level on routine chats to make your weekly limits last longer.

If you are deciding whether to pay at all: this makes Claude Pro a stronger $20 than it was yesterday, but it does not change the basic shape of the decision -- ChatGPT Plus is still the broadest, Google AI Pro still bundles the most, and Claude is still the writer's pick. We work through the whole comparison in which AI chatbot you should pay for.

The bottom line: Claude Opus 5 is the rare AI launch where the headline benefit lands on existing subscribers rather than on a new tier or a waitlist. Pro users got their best included model upgraded for free, five days after losing free access to Fable 5. Free users get nothing, and lose nothing.