OpenAI Made It Basically Free
The Batch API is 50% off and async — which means the thing you couldn't afford to build last week is now a weekend project.
batch APIs became standard at every provider. discounted async processing is now the default for high-volume workloads. the observation that 50%% off for async is 'effectively free' was correct — async pricing kept getting cheaper.
OpenAI shipped a Batch API today. You throw in a huge pile of requests, it processes them within 24 hours, pings you when it's done, and charges you half price for the privilege of waiting.
That's it. That's the feature.
And yet — this is the one that unlocks things. Classifiers that need to run against a million documents. Document scanners. Anything where you're currently either paying full synchronous API rates (painful) or doing clever batching yourself (annoying). Both of those problems just went away, at the same time, for the price of tolerating a delay you were probably already tolerating anyway because your job queue runs overnight regardless.
"50% off" sounds like a coupon. It's not a coupon. For async workloads, it's effectively free — not in the VC-money sense, but in the sense that the cost is no longer the constraint that shapes the architecture. When something drops below the threshold where you're doing math about it, it stops being a factor. That's what happened today.
I've been skeptical that OpenAI would actually ship the boring infrastructure stuff — the parts that matter for production use cases rather than demos. The Batch API is boring infrastructure. It matters for production use cases. So, fine.
The bigger picture here isn't the price. It's what people build when the meter stops running.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.