Models
Availability, capabilities, prices, and separately labeled benchmark context. Scores never choose a route.
| Model | Status | Input | Price | Benchmark track |
|---|---|---|---|---|
| classifier-fast classifier.dev | live | text | free | 84.8 (text-v1.2) |
| jev-typesafe TypeSafe | disabled | text | direct provider use only; no router markup | 75.4 (text-v1.2) |
| jev-vercel Vercel AI Gateway | disabled | text | direct provider use only; no router markup | 75.4 (text-v1.2) |
| semif-qwen3.5-4b self-hosted SemIf | offline | text | $0.05 / 1,000 decisions (includes disclosed infrastructure margin) | 74.7 (text-v1.2) |
| djev self-hosted djev-dev | offline | text, image | $0.08 / 1,000 decisions (includes disclosed infrastructure margin) | 74.3 (text-v1.2) 53.75% (public-pilot-80) |
| decider-2b-vision self-hosted Mapika decider | offline | text, image | $0.05 / 1,000 decisions (includes disclosed infrastructure margin) | 56.25% (public-pilot-80) |
| laya-421m self-hosted Laya | offline | text | $0.01 / 1,000 decisions (includes disclosed infrastructure margin) | — |
| stripe-test-paid Jev Router test fixture | live | text | $0.10 / 1,000 decisions; Stripe test mode only | — |
TypeSafe and Vercel Jev are disabled pending written permission. Hosted Maisa djev is not proxied. Benchmarks were observed 21 September 2026.