Tool-use agent over public APIs

Cohere command-a-plus-05-2026 with seven tools over five public APIs: Open-Meteo weather, Swiss public transport (transport.opendata.ch), Wikipedia, the GitHub REST API and ECB exchange rates (Frankfurter). Every model step, tool call, argument set and raw API result is shown below. Switch between the v1 and v2 tool descriptions to see the difference the eval measures.

Demo limits: 400 characters per question, 4 model steps, 5 questions per IP per 10 minutes, 8–10 s timeout per outbound API call. Runs on a Cohere trial key, so a daily cap applies.