Unify AI Providers with Vercel Gateway

This Promptfoo config compares three models from different providers, all accessed through the Vercel AI Gateway's unified API at 0% markup: OpenAI's gpt-4o-mini, Anthropic's claude-haiku-4.5, and Google's gemini-2.5-flash, each run at temperature 0.7 with a 200-token cap. All three answer the same simple-explanation prompt template across three topics: machine learning, quantum computing, and blockchain.

Each test uses a contains-any assertion checking for at least one expected keyword per topic - algorithm, data, learn, or pattern for machine learning; qubit, quantum, superposition, or computer for quantum computing; block, chain, ledger, or decentralized for blockchain - rather than requiring an exact phrase. Running the example requires a VERCEL_AI_GATEWAY_API_KEY environment variable.

Use it as a template for comparing models from different providers through Vercel's unified gateway, when you want single-API access across OpenAI, Anthropic, and Google without managing separate provider credentials. Do not use it if you're calling providers directly rather than through Vercel AI Gateway, or if you only need to test one model.

Input: the YAML config - a shared explanation prompt template, three vercel-prefixed provider configs, and three topic test cases with keyword assertions - plus the VERCEL_AI_GATEWAY_API_KEY environment variable. Output: Promptfoo's evaluation report comparing the three models' pass rates on the keyword-presence checks per topic.

Uses Promptfoo's vercel provider prefix to route calls to OpenAI, Anthropic, and Google models through the Vercel AI Gateway's unified API.

Teams standardizing multi-provider LLM access through Vercel AI Gateway who want to compare OpenAI, Anthropic, and Google models on simple-explanation quality.