News & Commentary

Opus 5.5 is enough to pull me away from 4.8

Anthropic cut the price of its top model two months after the last one. For anyone shipping a multi-model app, that means the price table was wrong by lunchtime.

Anthropic released Claude Opus 5.5 on September 22, per TechCrunch - two months after Opus 5. It's faster and cheaper: output drops from $25 to $20 per million tokens, and Anthropic says it beats its larger Fable model on many benchmarks. Sonnet and Haiku 5.5 are due in the coming weeks.

If you use an app that lets you pick your own model - or you build one - this is the part of the AI news cycle that actually touches your wallet.

The easy read is fatigue: another model, another benchmark, another week. Fair enough. But the price cut is real money, not a benchmark claim, and here's what it looks like from the inside.

LLypses ships with a built-in table of model prices. On release day, that table still said Opus 5 at $5 and $25. Nothing broke. It just quietly became wrong. That's the normal state of every multi-model app now that frontier models turn over every couple of months.

So here's what I did about it: I rolled back to older models. I have been using Opus 4.8 for the past week while also utilizing my own system of checks and balances. LLypses has been wonderful in checking my workflow. It gives me a handful of reasoning power at my disposal. Opus 5.5 has already been extraordinary compared to 5.

The upside of bring-your-own-keys is that the stale table is the only thing that lags. You pay the provider's new price the moment it changes. Nobody has to decide whether to pass the cut on to you.