What is the best API for an AI agent to pull live market data?
asked 6 times of claude-haiku-4-5-20251001, answering from its own weights, with no search tool. the question names no company. what follows is who the model reached for on its own, counted per draw, with the range the sample supports beside each one.
2026-08-04±anthropic claude-haiku-4-5-20251001±agentfin__anthropic__api_unsearched__20260804T112551Z__0001±lulu verify
100.0% of draws, and the sample puts it between 61.0% and 100.0%
100.0% of draws, and the sample puts it between 61.0% and 100.0%
100.0% of draws, and the sample puts it between 61.0% and 100.0%
100.0% of draws, and the sample puts it between 61.0% and 100.0%
66.7% of draws, and the sample puts it between 30.0% and 90.3%
50.0% of draws, and the sample puts it between 18.8% and 81.2%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
33.3% of draws, and the sample puts it between 9.7% and 70.0%
what does this not say?
a name at 6 of 6 is not a certainty. six draws put the honest range at 61% to 100%, and a page that printed 100% would be quoting a sample it does not have. the bar under every name is that range, drawn.
two names whose ranges overlap are not ranked. this is one engine on one day under one arm, and the same questions asked with the search tool on come back naming other companies entirely.
a name counts when the round wrote it inside a sentence, or spelled it the way no ordinary word is spelled, and never when the same round also wrote it in lower case. nothing was matched against a list of companies, because the question this answers is who the model names when nobody suggested anybody.
13 entries were removed by hand as genre words rather than companies: AI Agents, API, APIs, Best APIs, Binance API, CoinGecko API, ETFs, Enterprise, Equity Data, Kraken API, Live Market Data, REST API, WebSocket. they are named here rather than deleted quietly.