What Is Claude Opus 5? Anthropic’s New Opus Model, Explained

Claude Opus 5 is Anthropic’s newest Opus model, released on July 24, 2026, and it moves Opus 4.8 out of the top spot in the Opus line. The pitch is straightforward and aggressive: near-flagship capability at a workhorse price. Anthropic positions Opus 5 as its best-performing and most cost-effective everyday model, and says it delivers performance close to its top Mythos-class model on many tasks while costing far less. If you have been using Opus 4.8, this is the model that replaces it, at the same price, with a wider context window and more control over how hard it thinks.

The short version: Opus 5 keeps the $5 and $25 per million token pricing of Opus 4.8, adds a 1 million token context window and up to 128K output tokens, thinks adaptively by default across a five-level effort setting, and, on Anthropic’s own evaluations, outperforms Claude Fable 5 on coding and knowledge work. This piece explains what actually shipped, where Opus 5 sits in the Claude lineup, how it compares with Opus 4.8, and what the relentless release pace means for anyone building on these models.

What Anthropic shipped

Opus 5 is a capability and efficiency step rather than a rebrand. The headline specifications:

• 1 million token context window, a large jump that lets it hold a substantial codebase or a stack of long documents in a single session.
• Up to 128K output tokens, so it can produce long, structured responses in one pass.
• Adaptive thinking by default, with a five-level effort setting that ranges up to a maximum reasoning mode (the upper levels are named high, xhigh, and max). You dial the effort up for hard problems and down for routine ones to control cost and latency.
• Pricing held flat at $5 per million input tokens and $25 per million output tokens, the same rates as Opus 4.8.

Two further details matter for everyday users. Anthropic has made Opus 5 the new default model on Claude Max and the strongest model on Claude Pro, so many subscribers will be using it without opting in. It is also offered in a Fast mode that runs around 2.5 times the default speed, at twice the base price.

The effort control is the most practically useful change. Instead of a single fixed reasoning budget, you choose how much the model deliberates, which is a direct lever on both speed and spend. For a lot of everyday work you leave it low; for a thorny bug or a dense analysis you turn it up.

What Anthropic says it does not do well

A launch post that only repeats the wins is marketing, so it is worth recording the limits Anthropic states plainly. Opus 5 remains behind Mythos 5, the trusted-access model at the top of the range, on cybersecurity tasks. Anthropic notes that Opus 5 comes close to Mythos 5 at finding software vulnerabilities but is substantially behind at developing exploits for them, and says it deliberately avoided training Opus 5 on cyber tasks. In practice, requests flagged by the cyber classifiers fall back to Opus 4.8 by default in Claude.ai, Claude Code, and Claude Cowork.

On the safety side, Anthropic reports that its automated behavioral audit found Opus 5 to be its most aligned model to date, with the lowest rates of deceptive behavior among recent releases. That is also a vendor self-assessment, measured on Anthropic’s own audit, so it belongs in the same evidence category as the benchmark claims.

Where Opus 5 sits in the Claude lineup

To place Opus 5, it helps to remember the ladder. Anthropic’s Claude 5 lineup runs from the mid-tier Sonnet, up through the Opus workhorse tier, to the top Mythos-class models, Fable 5 and Mythos 5. Opus has always been the high-capability, general-purpose tier that most teams reach for when Sonnet is not quite enough and the flagship is overkill.

What makes Opus 5 notable is how far it closes the gap to the top. Anthropic describes it as delivering performance close to its most powerful model on many tasks at roughly half the price, and claims it outperforms Claude Fable 5 on coding and knowledge-work evaluations. Those are Anthropic’s own benchmark claims, so treat them as the vendor’s framing until independent testing confirms them. The takeaway that is not in dispute is the positioning: Opus 5 is meant to be the default model for enterprises, knowledge workers, and developers, the one you use for most real work rather than a premium you ration.

If you need help deciding which tier fits a given job, our guide to picking between Opus, Sonnet, and Haiku still applies; Opus 5 simply raises the ceiling of the Opus option.

How Opus 5 compares with Opus 4.8

Opus 4.8, released May 28, 2026, was the previous stable Opus and the model most Opus users were on until now. The move to Opus 5 brings a larger context window, the explicit five-level effort control, and Anthropic’s claimed capability gains, all while holding pricing steady at $5 and $25 per million tokens. In other words, you are meant to get more model for the same money.

For anyone who tracked the incremental Opus 4.7 to 4.8 step, Opus 5 is a bigger jump in name and in the context and effort features, though as always the real test is your own workload. Because the price did not change, the practical advice is simple: if you were on Opus 4.8, moving to Opus 5 is low-risk on cost, but re-run your own evaluations before assuming the benchmark gains show up in your specific tasks.

The relentless release pace

Opus 5 is, by Anthropic’s own count, its fourth Claude 5 model release in under two months. That cadence is the real story around the launch, and it is one we have written about directly in our piece on Anthropic’s faster Opus cycle. The pattern has shifted from occasional blockbuster launches to rapid, frequent improvements on capability, cost, and speed.

For teams building on these models, that pace cuts both ways. You get better and cheaper models sooner, but you also inherit more version churn: more re-testing, more prompt and evaluation maintenance, and more pressure to stay model-agnostic so a monthly release does not force a scramble. The upside is real, and so is the operational tax.

What it means for you

The competitive backdrop makes Opus 5 easy to read. Cheaper near-frontier models have been arriving from every direction, including OpenAI’s GPT-5.6 and a wave of capable open-weight models, and pushing a strong Opus down to a workhorse price is Anthropic’s answer to that pressure. For most organizations, the sensible move is to route everyday coding, analysis, and writing to Opus 5, reserve the top Mythos-class models for the genuinely hardest tasks, and use the effort setting to keep spend in check on high-volume work.

As with every model launch, the vendor’s benchmark claims are a starting point, not a verdict. Opus 5 looks like a meaningful upgrade at an unchanged price, which is a good deal on paper. Confirm it against your own workloads, watch how it holds up beside the cheaper alternatives, and remember that at this release pace, whatever you choose today is worth re-checking in a month. For a sense of how the cost math works once you start metering real usage, our explainer on input versus output tokens is a useful companion.

Frequently Asked Questions