Claude Opus 5 Explained: What Anthropic's New Model Does, What It Costs, and When to Use It

Claude Opus 5 launched July 24, 2026 at $5/$25 per million tokens, with near Fable 5 intelligence for half the price. Here is what it does and when to use it.

Claude Opus 5 Explained: What Anthropic's New Model Does, What It Costs, and When to Use It

Anthropic launched Claude Opus 5 on July 24, 2026. It runs at $5 per million input tokens and $25 per million output tokens, the same price as the model it replaces (Opus 4.8), and Anthropic says it comes close to the intelligence of its flagship Fable 5 model at about half the cost. For most business and marketing teams, that is the whole headline: near top-tier quality at mid-tier pricing. This article covers what Opus 5 is, what it costs, how it compares to the models around it, and where it actually fits in day-to-day work.

What Claude Opus 5 is

Claude Opus 5 is the newest model in Anthropic's Opus tier, the line built for hard reasoning, coding, and long-running agent work. Anthropic describes it as thoughtful and proactive, and positions it as a model you use every day rather than reach for only on the hardest problems. It is now the default model on the Claude Max plan and the strongest model available on Claude Pro.

A few specifics that matter in practice:

It is available today on all Anthropic platforms and through the Claude API as claude-opus-5.
Its knowledge cutoff is May 2026, the most current of any Claude model at launch.
It has a 1 million token context window, so it can hold large codebases, long transcripts, or big document sets in a single session.
It ships with an effort setting (low, medium, high, xhigh, and max) that lets you trade cost and speed for raw capability on a per-request basis.

The short version: Opus 5 is meant to be the workhorse. Anthropic built it so that the quality you used to pay flagship prices for is now the everyday tier.

What Claude Opus 5 costs

Pricing is the part most teams will care about first, and it is refreshingly simple. Opus 5 costs the same as Opus 4.8:

$5 per million input tokens
$25 per million output tokens

Anthropic did not raise the price for the better model. You get a meaningful capability jump at flat cost. On top of the base rate, two standard discounts apply: prompt caching can cut costs by up to 90% on repeated context, and batch processing (through the Message Batches API) cuts them by about 50% for work that does not need an immediate answer.

There is also a Fast mode that runs Opus 5 at roughly 2.5 times the default speed, offered at twice the base price on the Claude Platform and through usage credits in Claude Code. If latency matters more than token cost for a given workflow, that is the lever.

The effort setting is the other cost control, and it is worth understanding rather than ignoring. At low effort, Opus 5 spends fewer tokens and returns faster, cheaper answers. At max effort, it thinks longer and spends more to squeeze out the top of its ability. Anthropic's own data shows Opus 5 beating other models at a given cost across the high, xhigh, and max settings, so the practical move is to start lower and only raise effort on the tasks that actually need it.

How Claude Opus 5 compares to Opus 4.8 and Fable 5

The clearest way to read Opus 5 is as the model that closes most of the gap to Fable 5 without the Fable 5 price tag. Here is how the top Anthropic models stack up side by side.

--- | --- | --- | --- | --- |

Claude Opus 5 | $5 / $25 | Everyday reasoning, agents, long multi-step work | Near-flagship quality at half the cost, 1M token context, May 2026 cutoff, effort setting for cost control, most aligned model to date | Not the outright leader on every benchmark, deliberately weak on offensive cybersecurity tasks |

Claude Opus 4.8 | $5 / $25 | Prior-generation general work | Same price as Opus 5, still capable, used as the fallback when Opus 5 classifiers block a request | Roughly half the Frontier-Bench score of Opus 5, more tokens for the same result, older knowledge cutoff |

Claude Fable 5 | Flagship tier (roughly 2x Opus 5 per task) | Absolute peak intelligence on the hardest problems | Highest overall ceiling, still the top model on most coding and knowledge benchmarks | Roughly twice the cost per task for a small quality delta, harder to justify for routine work |

Claude Sonnet | $3 / $15 | High-volume everyday work where speed matters | Fast, cheap enough to run at scale, strong for drafting, summarizing, and routine coding | Noticeably weaker on long multi-step agent work and hard reasoning than Opus 5 |

Claude Mythos 5 | Specialist tier | Security research and exploit development | Strongest on cyber tasks, including turning found vulnerabilities into working exploits | Narrow use case, not the right default for marketing, ops, or general development |

OpenAI GPT-5.6 | Flagship tier | General reasoning, multimodal work, and the broadest tool ecosystem | Excellent all-rounder, deep integrations across the OpenAI platform, strong image and voice support | Costs more per task than Opus 5 for comparable knowledge work, and trails it on the long-horizon agent benchmarks Anthropic published |

How to choose: run Opus 5 as your default and raise the effort setting only on the hard tasks. Drop to Sonnet for high-volume drafting and routine coding, and to Haiku for simple classification or extraction at scale. Reach for Fable 5 when a job genuinely needs the top ceiling, keep Opus 4.8 only as a fallback, and use Mythos 5 solely for security work. If your stack already runs on OpenAI, GPT-5.6 is a fine general-purpose peer, but Opus 5 is the better value for long agent workflows and heavy knowledge work.

On coding and knowledge-work evaluations, Anthropic reports that Opus 5 is the new state of the art on benchmarks like Frontier-Bench and GDPval-AA. A few concrete numbers from the launch:

On Frontier-Bench v0.1, Opus 5 surpasses every other model and more than doubles Opus 4.8's score, at a lower cost per task.
On CursorBench 3.2 at max effort, it lands within 0.5% of Fable 5's peak score at half the cost per task.
On ARC-AGI 3, a test of solving genuinely novel problems, its score is three times the next-best model.
On Zapier's AutomationBench, which measures whether a model can finish real business tasks end to end, its pass rate is roughly 1.5 times the next-best model at the same cost per task.
On OSWorld 2.0, a computer-use benchmark, it beats Fable 5's best result at just over a third of the cost.

There is one honest limitation to note. Opus 5 remains behind Anthropic's Mythos 5 model on cybersecurity tasks, specifically on turning found vulnerabilities into working exploits. Anthropic says it intentionally avoided training Opus 5 on cyber tasks. For ordinary business, marketing, and development work, this does not matter. It only affects specialized security research.

Against Opus 4.8, the model it replaces, early-access customers reported gains across the board: better numerical reasoning and table work in financial analysis, cleaner and more concise code with fewer dead ends, and stronger performance on long, multi-step tasks. Several noted that Opus 5 hits similar or better results while using fewer tokens, which offsets some of its usage in real workflows.

What Claude Opus 5 means for marketing and operations teams

Benchmarks are useful, but the question that matters is whether this changes anything in your week. For teams doing marketing, operations, and content work, a few things stand out.

First, longer tasks hold together. Opus 5 is built for work that does not fit in one sitting, the kind of multi-step job where older models would lose the thread. One early customer, Zapier, described handing it a raw account-health spreadsheet and having it run a full churn-prevention sequence end to end: flagging at-risk accounts, alerting the right owner, and summarizing for the retention team, hitting a 100% pass rate where previous models failed. That pattern (take messy input, do the whole job, hand back something usable) is exactly the kind of work that fills a marketer's or operator's day.

Second, the judgment is better. Multiple testers described Opus 5 checking its own work, catching its own logical mistakes during planning rather than after, and pushing back when a proposed approach was flawed instead of going along with it. For anyone delegating real work, that reliability is worth more than a benchmark point. It means less rework and less babysitting.

Third, it produces stronger visual and document output. Testers highlighted better decks, cleaner formatting, and fewer slide errors on longer document work. If your team leans on AI to draft reports, one-pagers, or presentations, that is a direct quality gain.

For agencies and small teams specifically, the flat pricing plus the capability jump makes it easier to justify running repeatable operations through Opus 5: reporting, data cleanup, research summaries, and first-draft content. The effort setting means you can run the routine stuff cheap and reserve max effort for the work that needs it.

Safety and alignment

Anthropic says Opus 5 is its most aligned model to date. On its internal behavioral audit, Opus 5 scored the lowest of any recent model on misaligned behavior, showed the lowest rates of deceptive behavior, and was the least susceptible to being tricked into misuse. It also rated as the safest at avoiding reckless actions with hard-to-reverse consequences, which is the trait you want most in a model you are handing autonomy to.

On the security side, Opus 5's classifiers block a narrow set of cyber tasks like exploit generation and penetration testing, while still allowing legitimate uses like finding vulnerabilities in source code. When a request gets flagged in Claude apps, it falls back to Opus 4.8 by default, and developers can enable automatic fallbacks on the API so requests route to the best available model instead of being blocked outright.

Frequently asked questions

When was Claude Opus 5 released?

Anthropic released Claude Opus 5 on July 24, 2026. It was available the same day on all Anthropic platforms and through the Claude API.

How much does Claude Opus 5 cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens, the same as its predecessor Opus 4.8. Prompt caching can reduce cost by up to 90% on repeated context, and batch processing reduces it by about 50%. A Fast mode runs at roughly 2.5 times the speed for twice the base price.

Is Claude Opus 5 better than Fable 5?

Not overall. Fable 5 remains Anthropic's flagship. The point of Opus 5 is that it comes close to Fable 5's intelligence on coding and knowledge-work benchmarks at about half the cost, which makes it the better value for most everyday work. On some specific benchmarks like ARC-AGI 3, Opus 5 leads.

What is the effort setting in Claude Opus 5?

The effort setting (low, medium, high, xhigh, and max) controls how much the model thinks before answering. Lower effort is faster and cheaper; higher effort spends more tokens for stronger results. It lets you tune cost against quality per request rather than paying max on everything.

Should my business switch to Claude Opus 5?

If you already use Opus 4.8, the switch is straightforward: same price, better model. For teams doing agent workflows, long research or analysis tasks, content drafting, or operations automation, Opus 5 is a clear upgrade. For narrow cybersecurity research involving exploit development, Mythos 5 remains stronger.

What is Claude Opus 5's context window?

Claude Opus 5 has a 1 million token context window, large enough to hold big codebases, long meeting transcripts, or sizable document sets in a single conversation.

Bottom line

Claude Opus 5 is a straightforward win: near-flagship intelligence at the same price as the previous Opus model, with better judgment, stronger long-task reliability, and a knowledge cutoff of May 2026. For teams running AI agents, automating operations, or drafting content and reports, it is worth testing as your default and reserving higher effort settings for the work that needs them. The only real gap is specialized cybersecurity exploitation, where Mythos 5 still leads.

*Want help putting a model like Claude Opus 5 to work in your marketing and operations? Book a call with us.*

Claude Opus 5 Explained: What Anthropic's New Model Does, What It Costs, and When to Use It