Anthropic just dropped Claude Opus 5, and the pricing math alone made me sit up straighter. Five dollars per million input tokens. For context, that is the kind of number that makes you re-read the spec sheet because you assume there's a typo buried somewhere.

The model is outperforming Fable 5 on major benchmarks too, which is the part that really matters. Cheap tokens are great, but cheap tokens attached to a model that scores higher on benchmark evaluations is a different conversation entirely. You can read the full breakdown over at Geeky-Gadgets.

Let me put the pricing in developer terms. When you're building an app that pipes user input through a model for classification, summarization, then a response generation pass, your token costs stack up fast. Three model calls per user interaction can eat your margin before lunch depending on what you've been paying for inference. At $5 per million input tokens, the unit economics actually start working for products that aren't just enterprise contracts. You can build something real without pricing in a massive cost buffer.

The benchmark wins over Fable 5 are harder to dig into without the full technical report, but the headline scores suggest Anthropic has made genuine progress in overall model quality. These are results that matter if you're trying to ship software with an LLM in the loop, not synthetic numbers that look good in a marketing slide.

The timing is interesting too. Competition right now is aggressive, and Anthropic dropping a cheaper, stronger model puts pressure on everyone else. The token pricing pressure goes both directions. If $5 per million input tokens becomes the new baseline, companies charging premium rates for comparable or weaker inference are going to have a hard time justifying their numbers.

What I want to see next is the API documentation. Benchmarks tell you a model is smart. Latency curves, rate limits, and context window behavior tell you whether you can actually build with it. That is the stuff I'll be testing the moment I get sandbox access...