Claude Opus 5 Is Here: Near-Fable-5 Power at Half the Price

Anthropic released Claude Opus 5 on July 24, 2026, and the headline isn’t a huge intelligence jump — it’s price. The company says Opus 5 gets close to the capability of its top-tier Claude Fable 5 model on most public benchmarks while charging exactly half as much per token, and it adds a new low/medium/high effort dial that lets you trade speed and cost for reasoning depth on a per-request basis.

I’ve watched Anthropic’s release cadence closely this year, and this one reads differently than a typical flagship drop. Fable 5 is still the model Anthropic tells you to reach for on your hardest problems. Opus 5 is being positioned as the model you actually run all day, every day, because it no longer costs a premium to get most of the way there. That’s a meaningful shift in how a frontier lab is choosing to compete, and it’s happening right as Anthropic gears up for a widely reported IPO later this year.

What actually changed in Opus 5

Opus 5 is priced at $5 per million input tokens and $25 per million output tokens — unchanged from its predecessor, Opus 4.8. What’s different is what that price now buys. On most public benchmarks, Opus 5 matches or beats Fable 5, and it wins outright on two fronts Anthropic clearly wanted to highlight: agentic coding (Frontier-Bench) and novel reasoning (ARC-AGI-3). In agentic evaluations, Opus 5 posted 70.57% on OSWorld 2.0, 26.0% on AutomationBench, and 30.16% on ARC-AGI-3.

The model also picks up a 1-million-token context window as both its default and its maximum, with 128K max output tokens on standard requests, extendable to 300K through the Message Batches API using a beta header. For anyone running long documents, large codebases, or multi-step agent sessions through the API, that’s a real practical upgrade over just a benchmark bump.

The effort toggle: pay for exactly the thinking you need

The feature getting the most attention is the effort toggle — a per-request setting with three levels: low, medium, and high. Low effort is faster and cheaper, meant for routine requests where you don’t need the model to deliberate. High effort lets Opus 5 spend more time reasoning through a genuinely hard problem before it answers. Medium sits in between as the default balance.

This matters more than it might sound like on first read. Up to now, if you wanted a frontier-tier model’s best reasoning, you paid frontier-tier prices on every single call, whether the task needed it or not. The effort toggle lets a team route simple tickets, quick summaries, and boilerplate code through the same model at the cheap setting, then dial up to high only for the requests that actually justify it — on the same API key, same model, no need to juggle multiple model families to manage a budget.

Where it fits: Fable 5, Opus 4.8, and the rest of the lineup

Anthropic isn’t pretending Opus 5 replaces Fable 5. The company’s own framing is that Opus 5 “comes close” to Fable 5’s frontier intelligence at half the price, while still pointing users toward Fable 5 for the most advanced or highest-stakes projects. Opus 5 becomes the new default model on Claude Max, Anthropic’s premium consumer tier, and it’s the strongest model available on Claude Pro. Anthropic also specifically called it out as its most capable generally available model for scientific research, alongside the usual push on coding and long-running agent work.

That three-tier structure — Fable 5 at the top for maximum capability, Opus 5 as the new cost-efficient workhorse, and Sonnet 5 as the budget option with introductory pricing through August 31, 2026 — gives Anthropic a full price ladder to compete against OpenAI and Google on. It’s a similar playbook to what we covered when Sonnet 5 launched alongside Fable 5 earlier this year, just extended one tier further down in price while pushing capability up to meet it.

Why the timing lines up with an IPO push

Multiple outlets, including Bloomberg and VentureBeat, have tied this release directly to Anthropic’s preparations for a public offering later in 2026. A model that delivers near-flagship performance at half the cost is a straightforward story to tell investors: better unit economics, without giving up the “we’re still at the frontier” narrative that justifies premium pricing on Fable 5. It also follows the same pattern Anthropic used with Sonnet 5’s aggressive introductory pricing — compete on cost-per-token as hard as on raw capability, because enterprise customers increasingly buy on both axes at once.

That competitive pressure isn’t limited to Anthropic’s own lineup, either. It’s part of the same summer of open-weight and frontier releases we’ve been tracking — Google confirming Gemini 4 pretraining while shipping Gemini 3.6 Flash, and massive open-weight models like Kimi K3 closing the capability gap from the other direction. When a frontier lab cuts effective pricing in half on a flagship-adjacent model, it’s rarely happening in a vacuum.

What this means if you’re actually building on it

If you’re currently paying Fable 5 prices for workloads that don’t need Fable 5’s absolute ceiling, Opus 5 at half the cost with a high-effort mode available for the requests that do need extra reasoning is worth testing against your own evals before you assume you need to keep paying full price. If you’re on Claude Pro, you’re getting Anthropic’s strongest available model as your new default with no plan change required. And if your workload is genuinely at the frontier — the hardest coding problems, the most demanding research tasks — Anthropic is still telling you, in its own materials, to reach for Fable 5 rather than Opus 5.

FAQ

How much does Claude Opus 5 cost?

$5 per million input tokens and $25 per million output tokens — the same price as its predecessor, Opus 4.8, despite the capability increase.

Is Claude Opus 5 better than Claude Fable 5?

Not across the board. Anthropic says Opus 5 matches or beats Fable 5 on most public benchmarks and wins outright on agentic coding and novel reasoning tests, but the company still recommends Fable 5 for its most advanced and demanding projects.

What is the effort toggle in Opus 5?

A per-request setting with three levels — low, medium, and high — that controls how much reasoning the model applies before responding, letting you balance cost and speed against depth of reasoning without switching models.

Where can I use Claude Opus 5?

It’s the new default model on Claude Max and the strongest model available on Claude Pro, and it’s accessible via Anthropic’s API at the pricing above.