About Lev

Support tickets, emails and JSON records are the kind of input Lev is built to judge. Interfaze, a product of JigsawStack, Inc., published it on 24 September 2026 as an open-weight decision model, with a blog post, a Hugging Face repository created that day and the training code on GitHub a day later. Rather than write text, it answers typed questions and returns a probability for each option in a single forward pass.

Three kinds of question

A caller supplies some context and a set of questions: yes or no, pick one option from a list, or rate on a scale. Lev returns a probability for every allowed answer. Because it writes no text, it cannot reply with a label outside the list it was given, although it can still choose the wrong one from that list.

Compatible with TypeSafe's Jev

Lev is Interfaze's open counterpart to TypeSafe's Jev. It uses the same /v1/systemone request format, so an application written against TypeSafe's SDK can talk to Lev running on the customer's own hardware once the address is changed.

What it is underneath

Technically Lev is a LoRA adapter of about 170 MB on Alibaba's Qwen3.5-4B, released under the Apache 2.0 license. It runs on one GPU or, much more slowly, on a CPU. Interfaze's own comparison gives 68.9 percent across all 13 S1Bench subsets, against 76.1 percent for Jev in the same run, and 0.980 on the banking77 intent set. End-to-end latency on one H100 ranges from 414 to 654 milliseconds, against about 340 for Jev's hosted API. All of these are Interfaze's own measurements, and no outside evaluation of Lev had been found.

Where it fits

The case for Lev is routing, moderation, triage and intent detection at volume, using an open model a business hosts itself with no per-call charge. According to the model card, a few of the training datasets come with their own conditions, non-commercial licenses among them, so a business should read those before commercial use. Several unrelated projects also use the name Lev, so buyers should confirm they are using the interfaze-ai repository.

TrustList summary

Lev is an open model from Interfaze (JigsawStack, Inc.) that makes decisions, published on 24 September 2026. It answers typed questions about a support ticket, an email or a JSON record by returning a probability for each option, rather than free text.

Main features

Yes or no questions: returns the likelihood of a yes or a no for a given piece of context.
Pick-one questions: gives a probability for every listed option in a single pass.
Scale ratings: rates content on a set scale, with the same probability output.
Self-hosted use: runs on one GPU, or on a CPU much more slowly, under the Apache 2.0 licence.
Compatible requests: uses the same request format as TypeSafe's Jev, so code written for its SDK can point at a self-hosted Lev by changing the address.

Customers and markets

Teams that need routing, moderation, triage or intent detection at volume and want to host the model themselves. Some training datasets carry their own terms, including non-commercial licences, which should be reviewed before commercial use.

Key facts

Base model: Alibaba's Qwen3.5-4B, with a LoRA adapter of about 170 MB
Licence: Apache 2.0
Training code: published on GitHub a day after the model was released

Summary written by TrustList from the business's own published information, Nov 2026.

Benchmarks & AI stats

AI model rankings

Benchmark	Official	Community avg
No benchmark scores yet. Be the first to add one.

“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).

Sign in to add a benchmark score for this model.