autotrust/GEV-26B-Decide — new model trending #30 on Hugging Face

**95% of 60 random multi-step tasks completed** (shop, settings, mail; 3–7 clicks each) in **about 85 ms per click**:

the same success rate as JEV-27B-VL, 3× faster. The colour swatches carry no text, so that click is decided from the

screenshot alone.

**Robot arm: pick and place from a camera image.** At every step System 1 looks at the top camera image and answers two

questions: is the target left or right of the gripper, and above or below it? The arm moves accordingly and halves its

step whenever an answer flips. It grasps the cube, carries it and drops it in the tray (MuJoCo simulation).

<video src="https://huggingface.co/autotrust/GEV-26B-Decide/resolve/main/videos/robot_arm_pick_place.mp4" controls autoplay loop muted playsinline width="100%"></video>

61 ms per decision, so a whole pick and place takes 4–8 seconds of model time. It completed 40% of 20 random scenes:

close to the target its left/right answers are less precise than JEV-27B-VL's, so more grasps miss. Once grasped, 8 of 9

cubes ended in the tray.

Same scenes and tasks for every model in the family:

| | **GEV-26B-Decide** | [JEV-27B-VL](https://huggingface.co/autotrust/JEV-27B-VL) | [JEV-9B](https://huggingface.co/autotrust/JEV-9B) |

|---|---:|---:|---:|

| computer use: numbered boxes + element text (60 tasks) | 95% | 95% | 95% |

| time per click | **≈ 85 ms** | ≈ 260 ms | ≈ 200 ms |

| robot arm: pick and place (20 scenes) | 40% | **75%** | 50% |

| time per robot-arm decision | **61 ms** | 239 ms | 163 ms |

* For computer use, give it the element text (as an accessibility tree would). With the numbered boxes alone it

completes 15% and often declares the task complete too early.

* Use System 1 for simple visual questions inside a control loop. Asked to pick one of 8 motor commands directly, it

completed 0 of 10 scenes.

Demo code: [JEV-9B `vl/demos/`](https://huggingface.co/autotrust/JEV-9B/tree/main/vl/demos) (set `JEV_URL` to this

server). Per-episode results: [`reports/demos/`](reports/demos).

## Overview

**GEV-26B-Decide answers typed questions with a calibrated probability for every option; with thinking switched on, it

thinks only when it needs to.** System 1 decides in one forward pass (about 45 ms). When its leading option is uncertain, System 2 (the same

backbone in Gemma-4 thinking mode) reasons over the question, and the reasoning is folded into the final probabilities.

One set of weights, one vLLM engine, for text and images.

| | what it does | output |

|---|---|---|

| **System 1** | typed decisions: yes/no · pick one of 2–256 options · rate 0–5, over text and images; prompts up to 256K tokens | a calibrated probability for every option, in one forward pass |

| **Adaptive thinking** (opt-in: `thinking: "auto"`) | System 1 first; below 0.8 confidence, System 2 thinks and its answer is folded in | calibrated probabilities |

| **System 2** | the unmodified `google/gemma-4-26B-A4B-it`, optionally thinking step by step, text and images | text / reasoning |

GEV-26B-Decide was previously published as `autotrust/JEV-Gemma4-26B-A4B`; the weights are the same.

> **Two models, two organisations.** **TypeSafe Jev 1.13** is the hosted, closed model made by TypeSafe AI.

> **autotrust/GEV-26B-Decide** is an independent open-weights model built by AutoTrust AI; it is not affiliated with,

> endorsed by, or a product of TypeSafe AI.

## Watch it think

Each puzzle has one correct answer. System 1 answers in one pass; with `thinking: "auto"`, System 2 thinks when System 1

is below 0.8 confidence, and its answer is folded into the probabilities. The videos show puzzles that System 1 got wrong;

the tables below count all puzzles.

**Minesweeper.** Which hidden cell is certainly safe? System 1 is at chance (21.0 % against 25 %); with thinking, 86.0 %.

These thoughts usually reach the 8,192-token budget, and the answer read at that point is still right most of the time.

<video src="https://huggingface.co/autotrust/GEV-26B-Decide/resolve/main/videos/think_minesweeper.mp4" controls autoplay loop muted playsinline width="100%"></video>

**Connect Four.** Which column wins now, or stops the opponent from winning next move? 54.0 % → 99.5 %.

<video src="https://huggingface.co/autotrust/GEV-26B-Decide/resolve/main/videos/think_connect4.mp4" controls autoplay loop muted playsinline width="100%"></video>

**Wordle.** Which word still fits all the colour feedback? 52.0 % → 100 %.

<video src="https://huggingface.co/autotrust/GEV-26B-Decide/resolve/main/videos/think_wordle.mp4" controls autoplay loop muted playsinline width="100%"></video>

**Sudoku.** Which digit belongs in the highlighted cell? 76.0 % → 99.5 %.

<video src="https://huggingface.co/autotrust/GEV-26B-Decide/resolve/main/videos/think_sudoku.mp4" controls autoplay loop muted playsinline width="100%"></video>

One-move puzzles, 200 generated puzzles per game (threshold 0.8, budget 8,192 thinking tokens; text input, chess with the

board image as well):

| game | question (options) | chance | System 1 | **adaptive** | puzzles that thought |

|---|---|---:|---:|---:|---:|

| Minesweeper | which hidden cell is certainly safe (1 safe cell, 3 mines) | 25.0 | 21.0 | **86.0** | 100 % |

| Wordle | which word fits all the feedback (8 words) | 12.5 | 52.0 | **100.0** | 98 % |

| Connect Four | which column wins now or blocks (legal columns) | 14.5 | 54.0 | **99.5** | 95 % |

| Sudoku | which digit goes in the cell (1–9) | 11.1 | 76.0 | **99.5** | 92 % |

| 24 game | which expression equals 24 (6 expressions) | 16.7 | 89.0 | **100.0** | 74 % |

| Maze | first step towards the exit (2–4 directions) | 48.4 | 48.5 | **68.0** | 82 % |

| Chess (Lichess puzzles) | which move mates in one (16 moves, board image + FEN) | 6.3 | 46.5 | **78.0** | — |

Whole games and perception, where thinking helps little or not at all:

| check | System 1 | adaptive |

|---|---|---|

| Snake, one game (image + positions) | 3 food in 18 steps | 11 food in 90 steps (about 21 s per step) |

| Connect Four, 6 full games against a heuristic opponent | 1 win, 5 losses | 1 win, 4 losses, 1 draw |

| 2048, one game | 1,476 points | 1,016 points |

| Flappy Bird, one game | 0 pipes (23 frames) | 0 pipes (38 frames, about 35 s per frame) |

| Quick, Draw!, 320 real sketches, 16 answers (30 / 60 / 100 % of the strokes) | 46.9 / 73.1 / 94.4 % | 40.3 / 72.8 / 95.0 % |

Thinking pays off when the answer can be checked step by step against explicit rules: logic puzzles, tactics,

constraints, arithmetic. It does not help perception (sketches), reflexes (Flappy Bird) or long-horizon play (2048, full

Connect Four games), and every thought costs seconds. Details: `reports/thinking_games.json`.

## Adaptive thinking

1. **Fast distribution.** System 1 returns p1 in one pass.

2. **Think only when uncertain.** If the leading option of p1 is below the threshold (default 0.8), System 2 reasons in

Gemma-4's thinking mode over the same state, question and options. The reasoning length is yours to set, as with the

base model: `think_budget` caps the thinking tokens (default: no cap beyond the context window; the evaluations below

used 8,192).

3. **Fold the reasoning in.** When the thinking channel closes, the answer-letter distribution p2 is read in one step,

and the result is p = ½ p1 + ½ p2. On its own, p2 is close to one-hot and over-confident; the equal mix keeps the

reasoning's accuracy and System 1's calibration.

The threshold and the mix were chosen on 1,754 questions from six public sets that are not part of the Decision Index

(test or validation splits, 300 random questions each; AQuA-RAT has 254):

| set | System 1 | **adaptive** | thinking on | always think |

|---|---:|---:|---:|---:|

| AQuA-RAT (math word problems) | 68.1 | **89.0** | 53.9 % | 90.2 |

| LogiQA (logical reasoning) | 55.3 | **80.3** | 63.7 % | 82.0 |

| StrategyQA (multi-hop yes/no) | 68.0 | **78.7** | 71.0 % | 79.0 |

| MedMCQA (medical) | 66.7 | **71.7** | 52.3 % | 73.7 |

| OpenBookQA (science) | 94.3 | **96.3** | 14.7 % | 95.7 |

| CommonsenseQA | 86.7 | **85.0** | 32.0 % | 83.3 |

| **all 1,754** | 73.3 | **83.4** | 47.8 % | 83.8 |

Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/autotrust/GEV-26B-Decide