One model name that routes each coding request to the cheapest capable model
JevRouter by Straitly is a model router exposed as the single model name jev-router. It reads a whole coding conversation, picks the cheapest capable frontier model, keeps prompt caches warm, and returns 30% cashback as wallet credit.
JevRouter is a model router from Straitly that sits behind a single model name, jev-router. Instead of sending every coding request to one expensive model, it reads the request, judges what the work needs, and forwards it to the cheapest model in its pool that can handle it. The pool currently covers DeepSeek V4.1 Flash, GPT-6 Sol, Kimi K3, Astra, Opus 5.5 and Fable 5.1. On a blended index of Terminal-Bench 4.0, DeepSWE v1.1 and SWE-Atlas-QnA, Straitly reports a score of 65.3 against 65.99 for Claude Opus 5.5, at $7.82 per task versus $13.04. The routing method differs from the small keyword classifiers many auto-routers use. The conversation is cut into slices such as the task, pasted code, error messages and earlier turns. Each slice is read by its own instance of Jev, which scores what that part needs against benchmark results for every model in the pool. The instances then vote on one model. Because slices are read in parallel, the router can take in up to a million tokens of context, where single-pass routers typically look at the last message. It also tries to avoid switching models mid-task, so the provider's prompt cache stays warm. It is aimed at developers and teams who run coding agents and watch their model bill. Setup means pointing a tool at the Straitly base URL and setting the model to jev-router. Both the OpenAI and Anthropic APIs are supported, so tools such as Claude Code, Cursor, Codex, Cline and OpenCode work, with streaming and tool calls unchanged. Inline instructions like [jev: only use Opus] let a user pin or narrow the model choice, and the instruction is stripped before forwarding. An Activity page in the Straitly console shows which model answered each request and its cost. Among alternatives, JevRouter competes with static model selection and with generic LLM gateways. Its distinguishing points are conversation-wide reading, cache-aware switching, and a cashback scheme of 30% back as wallet credit on JevRouter and 24 frontier models. Cost and accuracy figures come from the vendor's own use of the Artificial Analysis Coding Agent Index.
Usage is billed per request at Straitly prices. 30% of spend on JevRouter and 24 frontier models returns as wallet credit, and other models return 5%. Spending $100 through JevRouter returns $30 usable on any model. No subscription tiers are stated.
Learn moreWorks with any tool that accepts a custom base URL and model name through the OpenAI or Anthropic API, including Claude Code, Cursor, Codex, Cline and OpenCode.
Cost-aware routing across six frontier models, parallel reading of up to 1M tokens, cache-aware switching, inline jev: steering, OpenAI and Anthropic API compatibility, streaming and tool call support, an Activity page showing model and cost, and cashback credit.
