OpenRouter Launches a Jev-Powered Cost-Aware Model Router

OpenRouter launched typesafe/jev-router, which uses TypeSafe AI's Jev, a small classifier-style model built to pick among a bounded set of options rather than generate free text, to choose which model and reasoning effort level handles each request instead of routing every call to one fixed model.

Developer Theo Browne independently tested jev-router on DeepSWE, a coding benchmark, and found it matched OpenAI's GPT-6 Astra at its lowest reasoning-effort tier on task success, but cost slightly more and took roughly five times longer to run. That is a narrower result than TypeSafe's own marketing suggests: Jev can be dramatically faster and cheaper than calling a large model directly on tasks that fit its bounded-choice format, a claim TypeSafe illustrates elsewhere with classification tasks, but routing an open-ended coding benchmark through it did not produce the same speed or cost advantage.