A curated, regularly-updated index of the best AI models, organised by what they are best at doing. Each collection gathers the models that are genuinely best at a specific use case, often because they were purpose-built or fine-tuned for it, with the quality signals that prove it and a clear sense of how much you save running them via Keld.
The frontier model isn't the right answer to every question. A 7-billion-parameter model fine-tuned for classification will tag your tickets faster and cheaper than a flagship, because that's the one thing it was trained to do. A speech model trained on thousands of hours of audio transcribes better than any general chat model. Choosing well isn't about settling for less; it's about picking the model that is the best at your specific job. This index is built to help you do exactly that: browse by collection, read the quality signals, and when you've found your fit, run it as a Keld Deadline job on the cheapest provider that serves it.
—
See exactly what you'd save with the Savings Calculator, or send any model in this index to the cheapest provider as a Keld Deadline job — max time, max price, delivered within your bounds.
Quality signals are indicative composites drawn from public evaluations and benchmark leaderboards; benchmark names denote the suite each model is commonly measured on. Savings are illustrative — the relative discount of a Keld Deadline job on the cheapest provider versus typical list pricing, refreshed periodically. "From $" figures are blended per-million-token starting rates and are not a quote.