OPUS 5 MATCHES FABLE 5 FOR HALF THE PRICE

Five dollars per million input tokens. Twenty five per million output. That is the whole story, because it is exactly what Opus 4.8 cost yesterday, and Anthropic is now claiming the model sitting behind those numbers comes close to Fable 5, which runs $10 and $50.

The capability moved. The price did not. In a year where every frontier lab has trained the market to expect the best model to cost more, that is the interesting part, and it is worth more attention than any benchmark chart.

$5 and $25, Unchanged From Opus 4.8

Anthropic released Claude Opus 5 on Friday, July 24, calling it "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price." The announcement post did 5.9 million views in its first few hours, which tells you the audience for a cheaper frontier model is not a niche.

The receipts, from Anthropic's own numbers: on coding work it more than doubles Opus 4.8 at lower cost, and lands within half a percent of Fable 5 on agentic coding tasks at half the price per task. On computer use it beats Fable 5's best result for just over a third of the cost. On abstract reasoning it scores three times the next best model. Those are vendor benchmarks, not independent tests, and they should be read as the company's best case. But the pricing is not a benchmark. That number is real on the invoice.

The Effort Dial Is the Feature Nobody Will Demo

The headline feature is an effort setting: low, medium, or high, chosen per task. High effort spends more tokens thinking. Low effort conserves them. You decide, per call, whether a job deserves the expensive version of the model.

This is the quiet feature, and it changes a habit rather than a spec sheet. Until now, picking a model meant guessing before you started. Now the same model can be cheap on the boring pass and expensive on the hard one, which is how anyone actually works. There is also a fast mode at twice the base price for roughly 2.5 times the speed, which is a straightforward trade rather than a trick. The practical effect is that the cost of a task stops being fixed by which model you picked and starts being a decision you make while working. That is a real behavior change.

The Same Company Spent July Yanking Fable 5 Around

Context matters here. Anthropic's actual frontier model has had a chaotic month: pulled offline entirely by export controls in June, then handed back to subscribers in a series of last minute extensions that kept moving the deadline. Fable 5 access was never something you could plan a workflow around.

Opus 5 answers that without saying it out loud. It becomes the default model on Max and the strongest model available on Pro, and it runs on Claude.ai, Claude Code, Cowork, and the API. Anthropic also says its safety classifiers should trigger about 85 percent less often than on Fable 5, which is the company admitting the frontier model was interrupting people. A model you can rely on at a price that does not move is a different product than a model that is briefly free and then gone.

The Lock In Score Is Low. Use That.

Here is the honest read on the trade. Switching costs on the API are close to zero, one model string, and the effort dial means you can dial spend down without changing models at all. Nothing about this is a trap. The cost is that you are calibrating your work to vendor benchmarks nobody outside Anthropic has replicated yet, and "close to Fable 5" is doing a lot of load bearing in that sentence. Close is not equal, and on the hardest problems the gap will show.

Try it, and specifically try it on the work you were rationing. If you have been sending cheap tasks to a weaker model to protect your budget, that habit is now probably obsolete, and the honest test is a week of real work rather than a benchmark you did not run. Anthropic is reportedly preparing an IPO later this year, which means the pressure to show revenue per token is about to get very public. Prices this good tend to be a moment, not a permanent condition. Use the moment.