Free AI decision models: where to use Jev free, plus Liquid d1 and other alternatives

Decision models are a new category that took off after TypeSafe released Jev in mid-September. They do not write text: you send a ticket, a conversation or some data, plus a few questions with possible answers, and the model returns a choice, a score or a yes/no probability your code can act on. For ticket routing, moderation or picking an agent's next tool, that is faster and cheaper than asking an LLM for JSON and parsing it.

Jev's official API charges $0.042 per million input tokens, and output is free. Start with free access or trial credits, then choose a paid service based on your usage.

Free ways to use Jev

1. OpenCode Zen: model id jev-1.13-free, called with an OpenCode Console API key at https://opencode.ai/inference/systemone/v1/systemone. It is free for a limited time with no end date announced. See the OpenCode Zen docs.
2. BeatAPI: a third-party reseller offering jev-1.14-free, usable with a zero balance. Before your first top-up each account gets 1 successful request per minute, and up to 10 after. See BeatAPI's Jev page.
3. The free Jev window on Vercel AI Gateway ended on September 25; see our earlier Vercel Jev page.

Free access and trial alternatives

1. Liquid d1 (Liquid AI): new organizations receive $2 in trial credits. The direct d1 API accepts text and images at $0.02 per million input tokens with free output; d1-ultra costs $0.042 per million input tokens. See Liquid pricing and the d1 docs.
2. Mercury Decide (Inception): inception/mercury-decide:free on OpenRouter is free and rate limited, with a 65,536-token context and the same request format as Jev. See the OpenRouter model page.
3. Span-01 Lite (Respan): the free model id is span-01-free. It scores behaviors in conversations, such as whether a user wants a human. There is a daily cap that resets at 00:00 UTC. See the Respan docs.
4. Alibaba Cloud Model Studio decision-model-preview: free for a limited time in the Beijing and Singapore regions. See our Model Studio decision model page.
5. vLLM-SR Decision 3.0: the site advertises a free API, with six sizes from 0.6B to 27B that read text, images and video. The free limits are not spelled out, so check the docs before heavy use.

Open models you can host yourself

1. Kev (Jared Palmer): an open Jev-style family in 0.8B, 4B, 9B and 27B sizes under Apache-2.0. It uses the same API as Jev, so switching means changing the endpoint and model name, and you can try it in a Hugging Face Space first. See the GitHub repo.
2. Laya (Convai Innovations): 421M parameters under Apache-2.0. pip install laya and it runs on CPU, with a 322M multilingual version covering 100+ languages. See the Hugging Face model page.

Which one to pick

To get a workflow running quickly, start with OpenCode Zen's jev-1.13-free, Mercury Decide Free on OpenRouter, or Liquid's new-organization trial credits. If your data cannot leave your network, host Kev or Laya yourself. Free routes change quickly, so bookmark this page; we will keep it updated.

See the AI decision model directory for more access options, current prices and setup instructions.