TypeSafe Jev Alternatives

Direct alternatives to TypeSafe Jev

TypeSafe Jev is a hosted decision model. You send application state and typed questions; it returns a choice, a score or a yes/no probability, each with a confidence estimate. It bills input tokens only, at $0.042 per 1M (per its docs, September 2026). People look for open alternatives when data cannot leave their own servers, when most of their content is not in English, or when they want to train on their own labelled examples. Jev's docs say it is not fine-tuned per customer and that its accuracy is best in English. The open projects below are a few weeks old; expect them to change quickly.

Open models you run yourself

• Laya: an Apache-2.0 decision model built on ModernBERT. It answers choice, score and yes/no questions across 100+ languages in one forward pass. The English checkpoint reads 512 tokens; the multilingual one reads up to 8,192. It installs with pip install laya and also ships a TypeScript package, an HTTP server, an MCP server and integrations for LangChain, LlamaIndex and CrewAI. A fit for multilingual workloads.
• Kev: Jared Palmer's Apache-2.0 family of Qwen-based decision models in 0.8B, 4B, 9B and 27B sizes. Its server matches TypeSafe's API: code written against the TypeSafe Python SDK can point at a self-hosted Kev unchanged. Its README also covers fine-tuning on your own labelled examples.

Hosted and open-weight options from October 2026

• Clef: Cloudflare's Apache-2.0 decision models in 27B and 9B sizes, hosted on Workers AI or run yourself. They accept Jev's request format, read images and video, and cost $0.24 or $0.09 per 1M input tokens on Workers AI.
• System1 Models: a German host of open decision models behind a Jev-compatible API, with an EU-only tier and a DPA. From $0.032 per 1M input tokens.

For a side-by-side comparison on price, context window and image input, see Decision models explained.