3 min read
Anthropic has released Claude Opus 5, its fourth flagship AI model in under two months, which is either an extraordinary run of engineering or a company that has forgotten the rest of us have to keep up. Mythos 5, Fable 5, Sonnet 5, and now Opus 5, all since roughly the start of June. At this rate Anthropic will run out of literary genres before the leaves turn.
What it actually is
Opus 5 is the big, expensive, top-of-the-range one, and by the benchmarks it has retaken the lead on the things enterprises actually pay for: coding, agentic work and computer use. It more than doubles its predecessor’s score on Frontier-Bench and reportedly triples the next-best model on the ARC-AGI 3 reasoning test. It comes with a one-million-token context window, enough to swallow a small library in a single prompt, a knowledge cutoff of May 2026, and a new higher-effort mode that lets you tell it to think harder when the problem deserves it. Cleverly, Anthropic held the price flat at $5 per million input tokens and $25 per million output, the same as the last Opus, while pitching near-Fable performance at roughly half the cost per task.
The treadmill
Here is the honest tension. Opus 5 is, by all accounts, genuinely excellent, and if you write code or wrangle documents for a living it is probably the best tool you can rent today. But the pace has gone slightly deranged. Four flagship launches in eight weeks means whatever you learned, integrated and optimised for last month is already a generation or two behind. The labs are locked in a benchmark arms race where the frontier lurches forward every fortnight, and the rest of us are expected to re-plumb our workflows on the same clock. “State of the art” now has the shelf life of a petrol-station sandwich.
Should you actually care
If you are a developer or a heavy knowledge worker, yes, this one earns a look, especially the effort control: pay for deep thinking when you need it, save the money when you do not. If you are a normal human who just wants a chatbot that works, the honest truth is you almost certainly cannot tell Opus 5 from the model it replaced, or from three of its rivals, and you should feel zero obligation to track which fashionable frontier model is briefly on top this week. The crown changes hands so often now that holding it for a fortnight is less a throne and more a hot potato.
Still, credit where it is due. Anthropic is turning out real quality at a punishing cadence, and Opus 5 looks like a genuine leap rather than a fresh coat of paint on last month’s model. Just do not act surprised when Opus 5.1, or Sonnet 6, or whatever comes next lands before you have finished reading the release notes for this one.
The AI tool stack actually worth paying for
One email a week. The tools, models and moves that matter, minus the hype and with the horseshit filter set to maximum. Free.
Did you know: a “context window” is how much text a model can hold in mind at once. A million tokens is roughly 750,000 words, or about nine copies of the first Harry Potter book, all readable by the model in a single prompt. The catch is that stuffing the window full does not guarantee the model pays equal attention to all of it, which is why “1M context” is a headline number, not a magic wand.