# Claude Opus 5: Overview, Price, and Whether to Switch in 2026
> A practical overview of Claude Opus 5 release claims, pricing, effort levels, and switching considerations. No independent quantitative benchmark comparison is provided.
Source: https://ailoot.ai/blog/en/claude-opus-5-review/
Canonical: https://ailoot.ai/blog/en/claude-opus-5-review/
Language: en
Publisher: Ailoot (https://ailoot.ai/)
Translation: guide-069
Revision: sha256-114712c50b62728ea43f8d832eaedbd35b91022f7d441d861bb75595d387e347
Key points
On July 24, 2026, Anthropic released Claude Opus 5, its new flagship replacing Opus 4.8. The main takeaway from the release: the model comes close to the most powerful Fable 5, but costs half as much, at $5 per 1M input tokens. The price stayed the same, while coding quality improved noticeably. It is already the default in Claude Code. This is a practical overview of the release claims, pricing, effort levels, and switching considerations. It does not provide a quantitative benchmark comparison or an independent evaluation.
For information on using the service, see the [Ailoot documentation](https://ailoot.ai/docs/).
## What Is Claude Opus 5, and Why Is Everyone Talking About It?
Key points
Claude Opus 5 is Anthropic’s new top-tier model, released on July 24, 2026, to replace Opus 4.8. Its positioning is this: flagship-level power, but you pay the price of a regular Opus. It became the default in Claude Code and the Claude Max subscription right away. If you build products through conversations with AI, that means one thing: you’re already working with the new model, even if you haven’t changed anything.
Six months ago, Anthropic rolled out Fable 5—a model that builds large projects almost from a single prompt, thoughtfully and neatly. Then they temporarily removed it from the subscription and offered it only for a high token price, and after that they temporarily disabled it altogether because of export restrictions. Now the story has come to a neat close: the same level of power is right in the subscription, and you pay what the previous Opus cost. That’s the news to me.
Here is how Anthropic describes the model itself:
A thoughtful and proactive model that approaches the frontier intelligence of Claude Fable 5—for half the price.
— Anthropic, https://www.anthropic.com/news/claude-opus-5
What this means in practice:
- **Opus 5 replaced Opus 4.8** as the top-tier model. If you were using Opus, you’ve already switched; there’s nothing to change.
- **It is the default in Claude Code** (starting with version v2.1.219) and in the Claude Max subscription ([release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.219)). On Claude Pro, it is the strongest available model.
- **Fable 5 is still around.** Anthropic explicitly says that for the hardest multi-day autonomous tasks, they still recommend Fable. Opus 5 is the workhorse for everyday use. Nobody is calling it a flagship killer.
In the work example described, Opus is the primary model for real tasks, so switching to Opus 5 is treated as a routine update to the everyday tool. Next, I’ll explain exactly what this update gives you.
## What Does the Release Claim About Quality?
Key points
Anthropic describes improvements over the previous Opus, but this article does not reproduce a named test with a harness, metric, and both scores. Treat this section as a summary of the release positioning, not a measured quality comparison.
A relative claim such as “more than doubles” is not interpretable here without the specific test and underlying values. No percentage quality gain is established by this overview.
I covered the Opus 4.8 comparison with Claude Code separately in the guide [“Opus 4.8 in Claude Code”](https://ailoot.ai/blog/en/opus-4-8-claude-code-features/)—it shows the baseline the model started from.
On working with interfaces (OSWorld), Opus 5 even outperforms Fable 5, at roughly a third of the cost, according to Anthropic.
To assess an upgrade for your project, compare the same task and acceptance checks on both models. Record completed work, failures, elapsed time, and total cost; this article supplies no such paired result.
## How Much Does Opus 5 Cost, and Why Is That the Main News?
Key points
The main news from the release isn’t the benchmarks, but the price. Opus 5 costs $5 per 1M input tokens and $25 per 1M output tokens—exactly the same as Opus 4.8. You pay what you did for the previous Opus, but get noticeably more for coding. In capability, the model comes close to the more expensive Fable 5. If you burn through tokens every day, this is the most tangible part of the release.
Opus 5 pricing:
| Item | Opus 5 | Opus 4.8 |
|-------------------------------------|-------------|------------|
| Input, per 1M tokens | $5 | $5 |
| Output, per 1M tokens | $25 | $25 |
| Context | 1M tokens | 1M tokens |
The price is confirmed by Anthropic itself: the same $5 and $25 as the previous Opus.
Price hits the wallet harder than benchmark numbers. Many people praised Fable 5, but it’s expensive, and it was sometimes removed from the subscription and sometimes brought back. Opus 5 addresses that pain: the same class of tasks, but at a price you can plan for. I covered a separate discipline for not burning through limits on any model in the guide [“How Not to Burn Through Claude Code Limits”](https://ailoot.ai/blog/en/claude-code-token-limits-saving/).
Choosing a model and effort level is one control. Context engineering is a more powerful one: what you put into the model’s context before the prompt affects the result more than a benchmark score does.
## What Are Effort Levels, and How Can You Save with Them?
Key points
Opus 5 can think at different depths, and you control that. Five effort levels are available in the API and Claude Code: low, medium, high, xhigh, and max, with high as the default. The higher the level, the longer and more expensively the model reasons, but the better it handles a difficult task. The app simplifies this to three buttons. The idea is simple: don’t run max for minor edits or set low for architecture—match the depth to the task and save tokens that way.
This is the most practical new control in the release. Before, the model thought as deeply as it decided it needed to. Now you set the depth with an effort parameter ([Anthropic documentation on effort](https://platform.claude.com/docs/en/build-with-claude/effort)). The Claude app has a simple toggle with three positions: low / medium / high. In Claude Code and through the API, there are five levels.
Here’s how I’d match them to a vibe coder’s tasks:
| Level | When to use it |
|----|----|
| `low` | minor edits, renaming, simple text—maximum savings |
| `medium` | ordinary coding tasks that don’t need deep planning |
| `high` (default) | complex code, architecture, analyzing a large project |
| `xhigh` / `max` | the hardest tasks, when mistakes are costly and you don’t mind spending tokens |
A practical routine to avoid overspending:
1. Start with `high`—it’s enough for almost everything.
2. For routine work, lower it to `medium` or `low`—the answer is faster and cheaper.
3. Raise it to `xhigh` or `max` only for genuinely difficult tasks: a major refactor, a confusing bug, architecture from scratch.
4. Remember: at `xhigh` and `max`, reasoning can’t be turned off—if you try, you’ll get an error.
One effort level won’t turn a bad prompt into a good one. But it really does save money: keep it at low for minor things, and your monthly bill gets noticeably smaller.
## Where Does Opus 5 Outperform Competitors, and Where Does It Lag?
Key points
Opus 5 hasn’t won outright: Anthropic still recommends Fable 5 for the hardest multi-day tasks. I’m not declaring it the “killer of all models”—it’s a strong general-purpose model with clear boundaries.
There’s no direct comparison between Opus 5 and OpenAI’s new agentic Codex yet: the models came out at different times and are tested using different harnesses, so nobody can currently put together a valid “Opus 5 versus Codex” table with exact figures. Anyone showing such a table has probably combined incomparable measurements.
Opus 5 has weaknesses too. Anthropic itself still recommends Fable 5 for the toughest multi-day autonomous tasks.
A separate note about Chinese open models like Kimi K3: there are many of them now, they’re cheap, and their benchmark results look good. I take them calmly and can say from experience: however good their numbers look, I have preferred Opus on my own tasks. That is a personal assessment, without a defined quality metric or controlled timing comparison. Claude has already finished the task while one of those models is still working on it. I covered this in more detail in the guide [“Kimi K3 vs. Claude Code”](https://ailoot.ai/blog/en/kimi-k3-vs-claude-code/). So Opus 5 wins on power and predictability; low cost isn’t its main selling point here.
## How Do You Enable Opus 5 in Claude Code?
Key points
Most likely, you don’t need to enable anything—Opus 5 is already the default. Starting with Claude Code v2.1.219, released on July 24, 2026, the default top-tier model is Opus 5 with a 1M-token context window. If you want to choose a model manually, use the /model command. Anthropic also cut about 80% of Claude Code’s system prompt for the new models—this changes how you should write your own CLAUDE.md files and skills.
Here’s what to do if you want to make sure you’re using Opus 5:
1. Update Claude Code to version v2.1.219 or later—that’s the version that made Opus 5 the default model ([GitHub release notes](https://github.com/anthropics/claude-code/releases/tag/v2.1.219)).
2. Open the model selector with the `/model` command.
3. Choose Claude Opus 5. In recent versions it’s already the default, so you can often skip this step.
4. Set the effort level if needed—`high` is the default; lower it for routine work.
One more detail for those who customize Claude Code. Anthropic cut about 80% of Claude Code’s system prompt for the new models.
According to Anthropic, coding quality didn’t drop. The point is that the new models need fewer hints in the system prompt—they already keep track of the task better. The takeaway for you is simple: don’t bury your agent in tons of instructions in `CLAUDE.md`. I keep mine short too—this works better with Opus 5 than long walls of rules.
## How Can You Check Regional Access Terms for Opus 5?
Key points
Also, since April 2026, Claude Pro and Max subscriptions can no longer be connected to third-party tools, so third-party clients require paid API access specifically. These are general, verifiable facts, with no instructions for bypassing sanctions.
A quick overview of access options:
1. **Claude Max or Pro subscription**—the simplest route through the interface; since April 4, 2026, these subscriptions no longer work with third-party clients.
2. **Paid API access**—if you use a third-party client on top of Claude, you need this specifically. A regular subscription won’t work here.
## Should You Switch to Opus 5?
Key points
The short answer: for most people, yes—and often you don’t even need to switch, because you’re already using it. If you were on Opus 4.8, this is a free upgrade at the same price; go for it right away. If you’re after maximum savings, use a low effort level or look at cheaper models. If you have the hardest multi-day autonomous tasks, Anthropic still recommends Fable 5. For everyday vibe coding, Opus 5 is currently the best balance of price and power.
Here’s how it breaks down by user type, so you can find yourself:
- **You use Opus 4.8.** Switch without hesitation. Same price, noticeably stronger on hard tasks, and already the default in Claude Code. This is a rare case where a model upgrade costs nothing.
- **You’re saving every token.** Opus 5 is a better value: keep `low` and `medium` for routine work and raise the effort only for difficult tasks. If your budget is really tight, compare it with cheaper models, but remember the difference in real-world work.
- **You have complex autonomous tasks that run for many hours.** Here Anthropic itself recommends Fable 5. Opus 5 is an excellent everyday model, but a different flagship is still the choice for heavy autonomy.
- **You’re just starting out.** Don’t worry about choosing a model at all. Use whatever’s set by default (that’s already Opus 5), and invest in how you prepare context. That matters more than the model itself.
I’ve been running Opus as my primary model for a long time, and every release repeats the same thing: a new model raises the ceiling, but it won’t build the project for me. What matters is what I put into the context; the benchmark score comes second. I’ve already set Opus 5 as the default and don’t think about it anymore. I think about how I prepare the project context in `business/`—that’s where the real leverage is. With or without Opus 5, it’s still up to you to build that structure.
Get to know [Ailoot](https://ailoot.ai/).
## Questions about Claude Opus 5: Overview, Price, and Whether to Switch in 2026
### Does this overview prove a numeric quality gain over Opus 4.8?
No. It summarizes the release positioning without a named benchmark, harness, and paired scores. Compare both models on the same project task and acceptance checks before drawing a quantitative conclusion.
### What should I record when comparing effort levels?
Use the same task and record whether the result passes your checks, the elapsed time, and total cost. Higher effort can spend more reasoning tokens; it does not guarantee a better result on every task.
## Ailoot · The company and its project portfolio
Ailoot combines established operating experience, a large client base, stable service and ongoing KPI control.
The company and its project portfolio: millions of customers; monthly sales in the billions.
Ailoot brings that operating experience to API access for AI features in customer products, agents, applications and automations.