For the past month, the frontier labs have been draped in sackcloth, publishing diatribes about slowing down, briefing the UN about existential peril (that they’re currently creating?), and generally behaving like people who’ve seen the future and need to go back and warn John Connor before it’s too late.
Then, on 22 September, Anthropic launched Claude Opus 5.5 and trimmed its price. Roughly ninety minutes later, OpenAI moved its Sol and Luna models to GPT-6 and cut the price in half. Two direct competitors, launching rival models and slashing prices within the same lunch break. Per people who track this stuff, that has genuinely never happened before. Doesn’t sound much like an industry tapping the brakes to me.
Anthropic dropped Opus from $5/$25 per million tokens to $4/$20, landing exactly on OpenAI’s price for the equivalent tier, to the cent. OpenAI’s response was to announce GPT-6 Sol at $2/$10, half the going rate, in about the time it takes to wolf down a sandwich after Anthropic’s move.
On the one benchmark both firms actually report, Opus 5.5 wins (40% versus 33.2%), so this is not necessarily OpenAI being cleaner or better, just cheaper, and betting most people care more about the smaller number on their never-ending subscription bill. The same executives warning that this technology might end civilisation are, it turns out, willing to race each other to the bottom on price to win your subscription this quarter. Talk of “slowing down” is lip service for the press release; never mind the bollocks.
For anyone actually paying the bills, the practical upshot is a good one: your AI costs are falling fast, and the sensible move is to re-price any project you shelved back when tokens were dear. Just don’t mistake a price war for a conscience.
Sources: WinBuzzer, Vellum, DataCamp, September 2026. Related: the slowdown that became a lawsuit.