autotrust/JEV-35B · Hugging Face

Source: https://huggingface.co/autotrust/JEV-35B

https://huggingface.co/autotrust/JEV-35B#%F0%9F%94%B4-live-autotrustgurusearch-recommendation–search-demo-%E2%86%92-newsguruso🔴 Live: AutoTrust/GuruSearch recommendation & search demo →news.guru.so

**Try AutoTrust/GuruSearch live (new, 11 October 2026).**A recommendation and search demo in which every ranking is a calibrated System 1 decision, made in about half a second: - **Recommendation:**the latest headlines from 6 news feeds, ranked by importance. - **Search:**results from Google News, Bing News and Yahoo News re-ranked by relevance, side by side with the search engines’ own order, plus a short answer with citations.

https://huggingface.co/autotrust/JEV-35B#a-system-1-decision-model-on-qwen35-35b-a3b-35-b-parameters-about-3-b-active-per-token-up-to-256-options-in-one-passA System 1 decision model on Qwen3.5-35B-A3B (35 B parameters, about 3 B active per token), up to 256 options in one pass

Decision Index 0.3, public suite: 59.73(our scoring with the kit), against 53.64 forautotrust/JEV-27B-VLon the board. Approximate public vision score71.94(our rebuild of the vision benchmarks; JEV-27B-VL 71.67 on the same rebuild, a tie within noise). Median single-request latency 204 ms on one B200 (JEV-27B-VL 271 ms).

NVFP4 version:autotrust/JEV-35B-NVFP4, the same model with its routed experts in NVFP4:25.6 GB download, 23 GiB of GPU memory(bf16: 72 GB, 66 GiB), so it runs on one 32–48 GB GPU. Decision Index 0.3 public 59.18 (bf16 59.73), vision rebuild 71.23 (bf16 71.94), 96.6 % of the Decision Index answers identical to bf16, same latency, sameserve\.shand API.

Two models, two organisations.****TypeSafe Jev 1.13is the hosted, closed model made by TypeSafe AI.autotrust/JEV-35Bis an independent open-weights model built by AutoTrust AI; it is not affiliated with, endorsed by, or a product of TypeSafe AI.

https://huggingface.co/autotrust/JEV-35B#decision-index-03-public-suiteDecision Index 0.3 (public suite)

public indexautotrust/JEV-35B, System 1(our run with the kit)59.73autotrust/JEV-27B-VL, System 1 (board, public part)53.64 area (skill)Knowledge & ReasoningLanguageRetrieval & ClassificationTools & AutomationArts & TasteJEV-35B0.4260.6320.7100.764****0.422JEV-27B-VL0.4140.5640.5370.7350.416 JEV-35B: all 140,178 scoreable requests of the 0.3 public suite answered (0 errors), System 1 only (thinking off), every choice read in one pass (up to 256 options), scored with the kit’sscore \-\-edition 0\.3. Our scoring, not a board entry: the board’s Full score also counts private tests (80 %), which only its maintainers run. JEV-27B-VL: the board’s public numbers for JEV-27B, whose text decisions JEV-27B-VL reproduces (see the JEV-27B-VL card).

JEV-35B is ahead in all five areas and on 24 of 37 benchmarks. Largest gains (skill points): PhishNChips +42.2, HoVer +28.4, Habermas Machine +28.1, VAST +22.7, WinoGrande +18.9, BANKING77 +18.3, iSarcasmEval +17.2, When2Call +14.9, GPQA +9.5. JEV-27B-VL is ahead on POP909-CL (+25.8), GSM8K (+18.2), ANLI (+11.4), BPoMP (+10.9), NLI4CT (+6.8) and BBH (+4.7).

https://huggingface.co/autotrust/JEV-35B#every-benchmarkEvery benchmark

Skill rescales the benchmark’s own metric so that chance is 0 (below chance counts as 0); ★ = gold benchmark (weight 1.2); the higher score is in bold.

Knowledge & Reasoning(area skill 0.426; JEV-27B-VL 0.414)

benchmarkJEV-35BJEV-27B-VLGPQA Diamond ★0.3600.265GSM8K (0.3 rebuild)0.3600.542ChessBench0.1270.091MuSR0.3610.402SATA-Bench0.2840.328CRUXEval0.5460.583CLadder0.4410.410HLE ★0.0000.000MMLU-Pro ★0.6420.567BBH ★0.6470.695WinoGrande ★0.8410.651 Language Understanding(area skill 0.632; JEV-27B-VL 0.564)

benchmarkJEV-35BJEV-27B-VLContractNLI0.7320.639ANLI ★0.5450.658HellaSwag ★0.9620.900ACOS0.2760.169FinEntity0.8550.755iSarcasmEval0.4260.254VAST0.5490.322NLI4CT0.6260.694RAGTruth0.6650.604 Retrieval & Classification(area skill 0.710; JEV-27B-VL 0.537)

benchmarkJEV-35BJEV-27B-VLBANKING77 ★0.9210.739CLINC150+OOS ★0.9430.845BRIGHT ★0.4210.417Amazon ESCI0.5200.430PhishNChips0.6600.238HoVer0.7620.478 Tools & Automation(area skill 0.764; JEV-27B-VL 0.735)

benchmarkJEV-35BJEV-27B-VLBFCL ★0.9500.946ToolRet0.6000.604API-Bank ★0.8560.825Home appliance simulator0.5000.523When2Call0.8620.714 Arts & Human Taste(area skill 0.422; JEV-27B-VL 0.416)

benchmarkJEV-35BJEV-27B-VLBPoMP0.7500.859Humicroedit0.2330.223POP909-CL0.1320.390cfcolor0.2560.283Habermas Machine0.4160.135New Yorker0.7440.607

https://huggingface.co/autotrust/JEV-35B#images-approximate-public-vision-scoreImages (approximate public vision score)

The vision benchmarks of the Decision Index are not published; this is our rebuild of their public datasets (CV-Bench, BLINK, RealWorldQA, CharXiv, InfographicVQA, Mind2Web, CORD + FUNSD, Hateful Memes, R-Bench-M, MMMU-Pro vision; Winoground not included), scored with the board’s chance correction and weights. The same rebuild gives JEV-27B-VL 71.67 against its board score of 71.53 on the same benchmarks.

approximate public vision scoreaccuracyECEJEV-35B71.9478.5 %0.050JEV-27B-VL (same rebuild)71.6778.0 %0.044 benchmarkJEV-35B skillJEV-27B-VL skillCV-Bench76.076.9BLINK55.856.7RealWorldQA68.372.2CharXiv79.780.7InfographicVQA95.095.9Mind2Web83.083.2KIE (CORD+FUNSD)98.698.6Moderation (Hateful Memes)48.438.0R-Bench-M28.130.3MMMU-Pro vision42.339.2 Overall the two models are level: the difference (+0.3) is inside the noise (paired bootstrap 95% interval −0.9 to +1.4). The only statistically significant per-benchmark difference (paired McNemar test, p < 0.01) is moderation, in favour of JEV-35B (+10.4); MMMU-Pro (+3.1, p = 0.03) also leans to JEV-35B. The small deficits on RealWorldQA, R-Bench-M, CharXiv, InfographicVQA, CV-Bench and BLINK (−1 to −4) are not significant. The vision tower is Qwen3.5-35B-A3B’s; image decisions are zero-shot.

https://huggingface.co/autotrust/JEV-35B#speedSpeed

One sequential client, the same 587 rows (387 text rows sampled across the Decision Index, 200 image rows), POST /v1/decide onserve\.sh(vLLM), one B200:

median (p95)textimageallJEV-35B184 ms (345)346 ms (744)204 ms (635)JEV-27B-VL (same benchmark)207 ms (617)457 ms (910)271 ms (746)

https://huggingface.co/autotrust/JEV-35B#computer-use-robot-arm-and-gamesComputer use, robot arm and games

Same demo code, seeds, scenes and opponents as theJEV-27B-VL card; every step is one System 1 decision (POST /v1/decide, thinking off), one model on one B200.

**Computer use: screenshot → which element to click.**A real browser (headless Chromium). Every clickable element gets a numbered box; System 1 picks the next click (or “the task is complete”), the browser clicks it, and the loop repeats.

Robot arm: pick and place from a camera image(MuJoCo). At every step System 1 answers two questions from the top camera: is the target left or right of the gripper, and above or below it? The arm halves its step whenever an answer flips.

JEV-35BJEV-27B-VLComputer use, numbered boxes + element text (60 tasks: shop, settings, mail)95%95%Computer use, numbered boxes only (60 tasks)**38%**10%ms per click decision, median (6 browsers in parallel)397720Robot arm, binary-decision servo (20 scenes): pick-and-place success75%75%Robot arm: median distance from the cube centre when grasping2.5 cm2.7 cmRobot arm: placed in the tray, once grasped15/1515/15Robot arm: ms per decision, median239239Robot arm, direct choice among 8 motor actions (10 scenes)0%0%

• Computer use with element text: 95%, the same three failures as JEV-27B-VL(shop seeds 12, 15, 20: the colour swatch carries no text and is skipped).
• **Numbered boxes only (every element read from pixels): 38% against 10%.**JEV-35B completes 90% of the settings tasks but only 15% of mail and 10% of shop; most failures still declare the task complete too early (24 of 37).
• **Robot arm: same success rate, different scenes.**Each model misses 5 of 20 grasps (both miss scenes 4 and 19), every miss 3 cm or more off the cube centre; once grasped, every cube reaches the tray.
• Choosing directly among 8 motor commands fails for both models; decompose control into simple visual questions.

Games(seed 0 for both models; board as image + text):

gameJEV-35BJEV-27B-VL2048: score / largest tile336 / 322,080 / 128Connect Four against a rule-based opponent (6 games)0 wins, 6 losses0 wins, 6 lossesFlappy Bird: pipes passed03Snake: food eaten1121Quick, Draw! top-1 among 16 (320 sketches; 30% / 60% / 100% of strokes)35.0 / 57.5 / 86.3%**41.6 / 62.2 / 87.8%**Chess mate in one, choice among 16 moves (200 Lichess puzzles)47.0%52.0% **For game play, use JEV-27B-VL.**Over five seeds JEV-35B’s largest 2048 tile averages 64 (random play: 102), it passes no Flappy Bird pipe and eats 3–18 pieces of food in Snake (mean 10). Per-episode results:reports/demos/.

https://huggingface.co/autotrust/JEV-35B#quick-start-vllmQuick start (vLLM)

hf download autotrust/JEV-35B --local-dir JEV-35B
bash JEV-35B/serve.sh          # vLLM on :8000; one GPU with 80 GB or more

On a smaller GPU, useautotrust/JEV-35B-NVFP4(one GPU with 32 GB or more; sameserve\.shand API):

hf download autotrust/JEV-35B-NVFP4 --local-dir JEV-35B-NVFP4
bash JEV-35B-NVFP4/serve.sh

serve\.shrunsserve\_decide\.py: the standard vLLM OpenAI server (same flags asvllm serve) with aPOST /v1/decideroute. The decision model is merged into the weights, so vLLM loads one plain checkpoint, with no LoRA. Tested with a vLLM development build from September 2026.

https://huggingface.co/autotrust/JEV-35B#system-1-post-v1decideSystem 1:POST /v1/decide

curl localhost:8000/v1/decide -H 'Content-Type: application/json' -d '{
  "kind": "choice",
  "state": "Customer message: my card was charged twice for the same order.",
  "question": "Which team should handle this ticket?",
  "options": ["billing", "shipping", "technical support", "account security"]}'

fieldvaluekind``noul: yes/no, probabilities for\["false", "true"\]·score: 0–5 ·choice: youroptions``statewhat the decision is about: a string, a JSON object, or a list mixing text and imagesquestionone question about the stateoptions``choiceonly: 2–256 strings The response hasoptions,probabilities,choice,choice\_indexandusage.GET /v1/decide/infolists the defaults.

If you call vLLM’s/v1/completionsyourself instead of/v1/decide, passtop\_k: 0andtop\_p: 1\.0so that the returned probabilities are not truncated, then add the head bias and apply the temperature as below. JEV-35B is a decision model only: it does not chat or generate text.

https://huggingface.co/autotrust/JEV-35B#engine-and-read-outEngine and read-out

System 1 reads the hidden state at the last token of a bare-text prompt:

[kind] choice
[state] ...
[question] ...
[options]
A) ...
B) ...
[decision]:

A 264-slot decision head gives one logit per slot; the active slots of the question’s kind are soft-maxed with a per-kind temperature (calibration\.json). The head is merged into thelm\_headrows of the slot tokens (A–JU,false/true,0–5);serve\_decide\.pyadds the head bias (adapter\_vllm/decision\_head\.json) and the temperature.

https://huggingface.co/autotrust/JEV-35B#limitationsLimitations

• A System 1 decision model only: no chat, no text generation, no thinking mode.
• Knowledge & Reasoning is the weak area (GPQA 0.360, MMLU-Pro 0.642, HLE below chance); JEV-27B-VL is better on GSM8K, ANLI and BBH.
• Weaker than JEV-27B-VL at sequential game play (2048, Flappy Bird, Snake).
• The vision score is our rebuild, not the board’s; image decisions are zero-shot.
• English-centric; not for high-stakes decisions without confidence gating.

https://huggingface.co/autotrust/JEV-35B#filesFiles

model.safetensors-* · config.json · tokenizer* · chat_template.jinja · generation_config.json · *_config.json
                          the decision model: Qwen3.5-35B-A3B architecture, decision head in the lm_head rows of the slot tokens
judge_config.json         slot layout, verbalizer ids, read-out
calibration.json          per-kind temperatures
adapter_vllm/decision_head.json
                          head bias and slot layout, applied by serve_decide.py
serve_decide.py · serve.sh
                          vLLM server with POST /v1/decide
videos/                   computer-use and robot-arm episodes
reports/demos/            per-episode computer-use, robot-arm and game results

NVFP4 weights (routed experts in NVFP4):autotrust/JEV-35B-NVFP4.

https://huggingface.co/autotrust/JEV-35B#licenseLicense

Apache-2.0. A fine-tune ofQwen/Qwen3.5-35B-A3B(Apache-2.0).