Over the past few months, my team and I have been building Blatent Struct. Today, we’re launching it. The idea was simple: an LLM should be fast, cheap and accurate. Of course, who doesn’t want that? But we also wanted to go beyond structured outputs and let you program the business logic that governs the answer. Structured outputs already let you constrain the JSON labels you get back. We wanted to express how fields depend on one another, what to loop over from the input or application state, and which paths the user’s permissions allow. With Struct, you define conditions, switch cases and loops, alongside constraints from OAuth scopes and OPA policies. Those rules are enforced during inference. Earlier decisions constrain what can come next, all within a single API call. So you can work through large trees of dependent decisions and get the complete JSON object back, without your application having to coordinate each step through separate model calls. The aim is dependable endpoints you can call throughout your software, with very little overhead. In our benchmarks, Struct achieves comparable accuracy to Claude Sonnet 5.5 at 11× the speed and over 100× lower estimated cost. It’s also over 2× faster than Jev, with higher average accuracy across 20 benchmarks. Across six extraction benchmarks, it scored 87.5% against Cloudflare Clairic’s 81.7%, at 62% lower estimated cost. I’m really proud of what our engineers and the wider team have built, and grateful to our investors and everyone who backed us. I hope you like it. I’m looking forward to seeing what you build. If you have questions, please reach out. It's free to get started (very cheap thereafter) and you can have 100m tokens on us :) My co-founder @R_Pisarczyk made an overview video here: xitter.cf/R_Pisarczyk/status/211… blatent.ai