Claude Opus 5 Released: Same Price, Much Stronger Coding

Anthropic released Claude Opus 5 on July 24. It sits in the same price bracket as the previous Opus model while closing most of the gap to their more expensive Claude Fable 5 in the work that matters day to day. (For a deep-dive on performance, check out our guide on 7 Tasks Claude Opus 5 Does Better Than Any AI).

The official post calls it a thoughtful and proactive model that comes close to Fable 5’s intelligence in many domains at half the token price. You can read the full announcement on Anthropic. Pricing stays at $5 per million input tokens and $25 per million output tokens. That is identical to Opus 4.8 and half of what Fable 5 costs. The context window is a flat 1 million tokens with up to 128k tokens of output on a normal request. Fast mode is available as a research preview; it roughly doubles the speed for double the price.

Availability is immediate across the Claude apps, the API (model ID claude-opus-5), Amazon Bedrock, Google Cloud, and Microsoft Foundry. On Claude Max, it is now the default. On Claude Pro, it is the strongest option you can select.

The numbers that stood out for me are the coding ones. On Frontier-Bench v0.1, which tests whether a model can turn engineering drawings into working software, Opus 5 more than doubles the score of Opus 4.8 and finishes ahead of every other model, including Fable 5. On CursorBench 3.2, at the highest effort setting, it lands within half a per cent of Fable 5 while using half the tokens per task. It also posts the best cost-to-performance numbers at the high, extra-high, and max effort levels. The Verge covered the same release and noted the enterprise push toward knowledge work and biology tasks.

ARC-AGI 3 is another one that jumped. Opus 5 scores roughly three times higher than the next-best model on novel problem-solving. Zapier’s AutomationBench, which checks whether models can finish real business workflows end to end, shows Opus 5 passing more tasks than any rival, even when you dial the effort all the way down.

What this means in practice is simple. For most of the coding and agent work I do refactoring modules, writing tests, debugging multi-file issues, or spinning up small internal tools Opus 5 is now the model I reach for first. It verifies its own output more carefully than earlier Opus versions and keeps going when the first attempt fails. In one of the evaluation examples Anthropic shared, the model was handed a machine-part drawing with no direct way to view it. It wrote its own computer-vision pipeline, extracted the geometry, and rebuilt the part in FreeCAD. No other model in the same setup managed it.

The effort control is useful. You can keep it on medium for routine work and only push it higher when the task is genuinely hard. That keeps token use in check without constant babysitting. Fast mode is there when you are in a tight edit-run loop, and the extra cost is worth the shorter wait.

Safety is tighter than previous Opus releases. Anthropic says this is the most aligned Opus they have shipped and the least likely to be tricked into misuse. It still sits behind Mythos and Fable on pure cybersecurity tasks, which is intentional. Guardrails fire less often than they do on Fable 5, so the day-to-day experience feels less interrupted.

I have been running a mix of real Next.js and TypeScript tasks through it since the day it dropped. The code it produces needs less cleanup. Long context holds up better when I drop entire folders in. It still makes mistakes, of course every model does but the iteration loop is shorter.

Fable 5 remains the better choice for the longest autonomous agent runs that stretch across days. For everything else that lives inside a normal workday, Opus 5 is the practical upgrade. Same price as the last Opus, noticeably stronger on the benchmarks that map to actual software work, and available right now.

If you already have a Claude Pro or Max subscription, switch the model selector and try it on the next ticket that would normally chew through a few hours of your time. The difference shows up quickly.