Decision modelOpen weights
AutoTrust JEV-27B-VL
AutoTrust AI
- Released
- September 30, 2026
- Data date
- October 7, 2026
AutoTrust JEV-27B-VL adds image processing to JEV-27B. System One scores text-and-image inputs with a probability for each option; System Two can reason step by step over images and text.
The decision head was trained on text only, so its image decisions are zero-shot and the model card says their calibration on image tasks has not been measured systematically. The underlying Qwen3.8-27B checkpoint and adapter are Apache-2.0; native context is 262,144 tokens.
Specifications and access
| Specification | Value and source |
|---|---|
| Model class | Vision variant with System One decisions and System Two text generationSource |
| Parameters | 27,781,427,952 in Safetensors metadataSource |
| Input modalities | Text and imagesSource |
| Output modalities | Decision probabilities or generated textSource |
| Base model | Qwen/Qwen3.8-27B with a vision towerSource |
| Image training | The decision head was trained on text only; image decisions are zero-shot. The System Two path can process images.Source |
| Context window | 262,144 tokens natively; the full length must be enabled explicitly in vLLMSource |
| License | Apache-2.0Source |
Page 1 of 7
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
CV-Bench · public score (Show measurement, test conditions, and source)
- Source value
- 76.61
- Score
- 76.61
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
BLINK · public score (Show measurement, test conditions, and source)
- Source value
- 62.95
- Score
- 62.95
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
RealWorldQA · public score (Show measurement, test conditions, and source)
- Source value
- 71.98
- Score
- 71.98
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
CharXiv · public score (Show measurement, test conditions, and source)
- Source value
- 79.7
- Score
- 79.7
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
InfographicVQA · public score (Show measurement, test conditions, and source)
- Source value
- 98.36
- Score
- 98.36
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
Mind2Web · public score (Show measurement, test conditions, and source)
- Source value
- 78.62
- Score
- 78.62
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
Winoground · public score (Show measurement, test conditions, and source)
- Source value
- 83.75
- Score
- 83.75
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
KIE (CORD+FUNSD) · public score (Show measurement, test conditions, and source)
- Source value
- 91.99
- Score
- 91.99
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
Moderation (Hateful Memes) · public score (Show measurement, test conditions, and source)
- Source value
- 40.31
- Score
- 40.31
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
R-Bench-M · public score (Show measurement, test conditions, and source)
- Source value
- 30.7
- Score
- 30.7
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
MMMU-Pro vision · public score (Show measurement, test conditions, and source)
- Source value
- 40.07
- Score
- 40.07
- Metric
- public vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=public. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
CV-Bench · private score (Show measurement, test conditions, and source)
- Source value
- 42.04
- Score
- 42.04
- Metric
- private vision benchmark score
- Unit
- %
- Benchmark version
- Decision Index 0.3 vision suite
- Category
- decision-index-vision
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Jev Decision Index 0.3 vision-results data
- Status
- active
- Retrieved at
- 2026-10-07
- Methodology
- Vendor-reported result. Jev Decision Index 0.3 vision System 1 evaluation for exact model https://huggingface.co/autotrust/JEV-27B-VL; column=private. The JSON file does not publish exact image preprocessing or per-task sample counts.
- Context
- Ran through the author's own server (vLLM + POST /v1/decide, System 1 only) with the context cap lifted to 256K; its base was read through a different (head-free autojev) path.
Page 1 of 7
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 7, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Additional source | AutoTrust JEV-27B-VL announcement (retrieved October 7, 2026) · AutoTrust JEV-27B-VL announcement · Editorial description reviewed October 7, 2026 |
| Additional source | AutoTrust JEV-27B-VL model card (retrieved October 7, 2026) · AutoTrust JEV-27B-VL model card · Editorial description reviewed October 7, 2026 |
| Additional source | AutoTrust JEV-27B-VL weight metadata (retrieved October 7, 2026) |
| Additional source | Jev Decision Index 0.3 vision-results data (retrieved October 7, 2026) |
| Additional source | AutoTrust JEV-27B-VL model card (retrieved October 7, 2026) |
| Additional source | AutoTrust JEV-27B model card (retrieved October 7, 2026) |