OpenAI's 80% Price Cut Wasn't Generosity. It Was Competition
OpenAI cut GPT-5.6 Luna prices by up to 80%, citing efficiency gains. But growing competition from Anthropic and Chinese AI labs like DeepSeek may explain why prices are falling now.
So OpenAI just cut prices on two of its GPT-5.6 models, Luna dropped a lot, something like 80%, Terra a smaller amount, and the official line was something about “advancing both capability and efficiency so each generation can accomplish more work at a lower cost.” Which is the kind of sentence a press release generates when nobody wants to say the actual reason out loud.
Here’s my honest take, opinion, not something I can prove. I don’t think OpenAI or Anthropic cut prices this aggressively if Chinese models hadn’t shown up and gotten actually good. If it was just the two of them and Google sitting comfortably ahead of everyone else, I have a real suspicion pricing would’ve stayed a lot cozier for a lot longer. Not saying anyone sat in a room and agreed to fix prices, zero evidence of that, wouldn’t say it if I had it, that’s a specific and serious legal claim. I’m saying the incentive to actually compete on price wasn’t really there until it suddenly was, and the timing tells you something.
DeepSeek dropped R1 back in January 2025, matched OpenAI’s o1 on math benchmarks while charging way less per token, something like 27 times cheaper if I remember the ratio right. Sam Altman posted on Twitter that same week calling it an impressive model, particularly for what they deliver at that price. Nvidia lost close to 600 billion in market cap the next day, one of the largest single day losses any company’s ever had. That’s not a small ripple. That’s the moment the industry realized the price floor everyone assumed existed, didn’t actually exist.
It’s basically been one long slide since then. Chinese labs have been shipping new flagship models way more often than the big three US labs combined, roughly double the pace from what I’ve seen tracked. And usage numbers back it up too, a decent chunk of US company AI spend has been quietly shifting onto Chinese models this year, not a trickle, a real and growing share. One startup, Lindy, apparently moved all of their traffic off Claude and onto DeepSeek back in June, their CEO described watching the cost curve just crash, said it’d save them millions.
OpenAI and Anthropic’s price moves didn’t happen in a vacuum pointed only at China either, they were pointed at each other at the same time. Back in June there were reports OpenAI was weighing cuts partly because it expected Anthropic to cut first, and Anthropic’s enterprise revenue had exploded that year mostly off Claude Code, eating into OpenAI’s own market share. So you had OpenAI watching Anthropic take share, both of them watching Chinese models undercut them hard on cost, everyone moving around the same few months. Feels less like coincidence and more like what happens when a price floor everyone was comfortable at finally cracks and nobody wants to be the one still charging premium a quarter later.
The actual cost gap between the US labs and the Chinese ones is genuinely big too, running the same workload through Claude or ChatGPT costs multiple times what running it through DeepSeek or Kimi or GLM does, we’re talking a real order of magnitude difference in a lot of comparisons I’ve seen floating around, not some small efficiency gap you shave off with a minor optimization.
Do I think it’s illegal price collusion in the legal sense, an actual antitrust thing? No, and I want to be clear I’m not accusing anyone of that specifically, no evidence for it and it’s too serious a claim to throw around loosely. What I mean, informally, is more like tacit price discipline, the kind that just naturally happens when three or four dominant players don’t need to talk to each other to all land on similarly high prices, because none of them have a reason to blink first. Not illegal. Just how markets like this tend to sit until something forces somebody to move.
The counter argument’s not nothing either honestly. Model efficiency really has improved a lot industry wide, compute costs per token have genuinely dropped from actual hardware and software gains, that part’s real, not just a cover story. And OpenAI versus Anthropic competing with each other might’ve eventually pushed prices down some on its own too, without China ever entering the picture, just slower probably. Can’t prove China was the whole reason. Don’t buy that it was a minor factor either, given how the timing lines up.
Cheaper AI is good for basically everyone using it so I’m not complaining about where this landed. Just don’t think anyone should pretend it was generosity.