Elektric – Benchmark your AI requests across 150 models for cost/quality (elektric.ai)

I built Elektric to solve a widespread problem in AI today: people pick models based on vibes, not data.

Elektric Benchmark collects 1000 representative requests from your AI app and evaluates them across 150 models to compare your current setup on quality and cost. The key difference from public benchmarks is that Elektric compares your actual AI traffic.

From there, Elektric can do two things with that information.

Elektric Grid is our automatic router. Your AI app sends us a request and we pick which model should handle it based on the request's characteristics and difficulty. Simply, not all requests need a frontier model. Many requests can be answered by capable but cheaper models like Luna. To use Grid, you simply change the base URL and API key of your OpenAI-compatible SDK, or use our SDK, and begin sending requests with "model: elektric-grid".

Elektric Wire - users that want to choose their own model can access 150+ models from all leading providers through one API and one account balance without maintaining integrations with multiple providers.

Some useful links: - The AI Model Should Disappear: https://elektric.ai/blog/the-ai-model-should-disappear - Why We Built Elektric on Cloudflare: https://elektric.ai/blog/why-we-built-elektric-on-cloudflare - How Elektric Determines Quality for Benchmarks: https://elektric.ai/blog/how-elektric-determines-quality-for...

I am interested in people trying it and telling me where it breaks, whether something is confusing, and most importantly, how useful is the benchmark idea actually. All questions and comments are welcome here or on vasileios@elektric.ai