Tool profile · Checked September 30, 2026
AutoTrust JEV-27B: Open Weights and Typed Decisions
AutoTrust JEV-27B is an open-weight model for developers who want to operate typed decision inference and text generation themselves. It is separate from the hosted TypeSafe Jev service.
Source-based editorial profile. No account, model inference or paid workflow was tested.
Two names, two organizations
AutoTrust’s model card explicitly disclaims TypeSafe affiliation, endorsement, shared weights or shared code. It describes a student trained using a corpus of TypeSafe output distributions. It is not an official open-source TypeSafe release or a successor to that company’s model.
Read the TypeSafe Jev hosted API profile for the alternative deployment choice.
Sources: Autotrust. Checked September 30, 2026.
Model and inference paths
The release is Apache-2.0 and based on Qwen3.8-27B. The card lists a 25.6B text backbone plus a 108.9M trained decision block. System 1 supplies typed noul, choice and score probability outputs; System 2 supplies generation and reasoning. AutoTrust publishes calibration data and a vLLM route that selects the decision adapter per request.
Sources: Autotrust. Checked September 30, 2026.
The repository contains downloadable weights and deployment files. Self-hosting transfers serving cost and maintenance to the operator; open weights do not mean free inference.
Sources: Autotrust Files. Checked September 30, 2026.
How to read the reported results
AutoTrust reports decision benchmarks and calibration measurements. Those results are not an AI Tool Finder performance winner or independent reproduction. Test accuracy, calibration and latency on your own examples before choosing thresholds.
Sources: Autotrust. Checked September 30, 2026.
A probability becomes useful only in relation to the decision it controls. For a ticket-routing trial, keep a held-out labeled set, include ambiguous cases, and compare the model with the existing routing rule. Measure abstentions and errors separately rather than treating every returned score as reliable confidence.
Hosted API or self-managed model?
Choose self-management when control over the inference environment justifies maintaining the runtime, capacity, upgrades and monitoring. Consider a hosted API when you prefer to call a service without operating the model. Neither route removes the need to validate decisions in your workflow.
- Inspect the pinned model files, license and official serving instructions.
- Validate both the decision route and generation route separately.
- Use labeled examples to set a human-review threshold before automating actions.
We have not downloaded the weights or run inference. Hardware suitability for your workload, production throughput and total operating cost remain unverified.