Anthropic put out Claude Opus 5 on Friday 24 July, and the interesting bit is not another benchmark chart. It is the price tag, or rather the fact that they did not move it.
Opus 5 costs $5 per million input tokens and $25 per million output tokens. That is exactly what the previous Opus cost. (A token is roughly three quarters of a word, and the meter runs on everything you send in and everything the model sends back.) So you get a brand new flagship at last year’s flagship rate. In a market where every vendor loves a bit of fee creep, holding the line is worth a mention.
What you are actually getting
Anthropic’s pitch is that Opus 5 comes close to the smarts of Claude Fable 5, its most capable public model, for half of what Fable 5 charges. Independent testers back the direction of travel. Artificial Analysis put Opus 5 top of its Intelligence Index at 61 on launch day, and top of its Agentic Index (the one that measures how well a model runs multi-step jobs on its own) at 55.3.
The focus this time is agentic coding and computer use, the buzzword for a model driving a browser and desktop like a person would, clicking and typing its way through a task. Context runs to a full million tokens as standard, and there is no smaller cut-price variant to upsell you later. Opus 5 is now the default on Claude Max and the strongest option on Claude Pro. The model ID, if you are wiring it into anything, is claude-opus-5.
The catch, and who it is for
Half of Fable 5 is still not cheap. $25 per million output tokens is real money once an agent is grinding away for hours, and ‘close to Fable 5’ is Anthropic’s own verdict, not a neutral one. If your work is short prompts and quick answers, a Sonnet-class model or one of the cheap Chinese open-weight models will do the job for a fraction. Opus 5 earns its keep on the hard stuff: long coding runs, thorny reasoning, agents that need to stay on task without wandering off.
The bigger picture is a price war finally reaching the top shelf. Frontier intelligence used to mean paying frontier rates. Anthropic just kept a frontier model at the same price for a second year running, and put it in front of ordinary Pro and Max subscribers rather than fencing it off for enterprise buyers. For anyone building on this stuff without a corporate card, that is the part that matters.
Did you know: the word ‘agentic’ barely existed in mainstream tech writing three years ago, and now it is the single most fought-over word on every AI pricing page.