TypeSafe: Jev Router

TypeSafe: Jev Router is a text model from TypeSafe, listed under the model id typesafe/jev-router. According to the provider, it picks the best model and reasoning effort for each request, balancing quality, speed, and cost. It runs on Jev, described as TypeSafe's first System One model, and adapts as your usage evolves. The provider release date is 2026-09-25.

The model accepts audio, file, image, text, and video as inputs, and it produces text as its only output. Its context window is 1,000,000 tokens, which is large enough to hold very long documents, transcripts, or multi file inputs in a single request. Reasoning is exposed as a supported parameter, and the model also supports include_reasoning, so you can decide whether reasoning content is returned alongside the answer. No maximum output token figure is published in the model details, so treat output length as something to control through your request parameters rather than a documented ceiling.

Because it takes audio, image, video, file, and text inputs and returns text, the model fits tasks that mix media with language. Examples include summarising recorded meetings or calls from audio, extracting structured information from scanned documents or PDFs, describing or answering questions about images and video frames, and running question answering over large text corpora that fit inside the context window. The router behaviour described by the provider means the same model id can serve both quick, low effort requests and harder requests that need more reasoning, without you switching between model ids.

You call it through api14's OpenAI-compatible API using the model id typesafe/jev-router. The request parameters it supports are frequency_penalty, include_reasoning, logit_bias, logprobs, max_completion_tokens, max_tokens, min_p, parallel_tool_calls, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_a, top_k, top_logprobs, top_p, and verbosity. Tool use is covered through tools, tool_choice, and parallel_tool_calls, and structured_outputs plus response_format let you constrain the shape of the returned text. For reproducibility you can set seed, and for sampling control you have temperature, top_p, top_k, min_p, top_a, and the penalty parameters.

Since the model routes internally, the exact model and reasoning effort behind a given response may vary, which is worth knowing if you need fully deterministic behaviour across runs. In that case, a fixed model with a single reasoning setting may suit you better. Otherwise, typesafe/jev-router is a reasonable default when your inputs span text, audio, image, video, and files, and you want one model id to handle both simple and reasoning heavy requests.