hjLabs AI Playground
Starting the on-device runtime…
A hosted API gives you frontier-quality models with no download and no hardware requirement, billed per token. Browser inference gives you zero marginal cost, no rate limits, no key management, no data egress and offline operation — at the price of a one-time download and a hard ceiling on model size somewhere around 1–3 billion parameters. This page maps out which workloads fall on which side of that line, including the cases where the browser is not merely cheaper but the only compliant option.
browser ai vs apiopenai api alternative freelocal inference vs cloudai cost comparisonon device ai privacygdpr ai processingno api key ai toolsedge inference tradeoffs
Long-context reasoning, code generation, anything needing a frontier model, and any workload where your users are on low-end phones. A 1.7B model in a browser will not replace a frontier model.
High-volume, low-complexity work — embeddings, classification, transcription, background removal — and anything involving data you are not allowed to transmit. Those are exactly the tasks where API bills accumulate fastest.