Developed by iApp, launched with AIEAT. Part of OpenThai (formerly OpenThaiGPT).
The vision variant of OpenThai-SystemOne: the same open Thai + English
System One decision model, now with images in the state and a fourth question type, point, that returns where
on a screenshot something is. Screenshot in, click out, in one forward pass, no text generation.
v1.5 (2026-11-20) learns to click in desktop apps the way an app with accessibility access does: given the list of
controls on the screen (Windows, macOS or Linux) it picks the one a Thai or English command means and answers none when that
control is not there. It also points better on Windows 11 and Ubuntu screens in any look (light / dark, display scale, full
screen) and reads "click + a name on the screen" voice commands right. On public screenshots with zoom it is still a little
behind v1.4 (see Evaluation).
v1.4 (2026-09-28) clicks better on Windows apps (Settings, Calculator) and on icons, and understands voice commands a little
better; it plays Doom from pixels worse than v1.3 (see Evaluation).
v1.3 (2026-09-26) clicked better on real Windows and Mac apps (trained on live Windows 11 and macOS screens) and on
large screens, and adds point_refine;
v1.1 (2026-09-24) added Thai voice-command understanding for hands-free computer control (a spoken Thai command +
the app and window title in, the operation and its target out) and stronger desktop and Doom play. See Versions.
| OpenThai-SystemOne (text) | OpenThai-SystemOne-Vision | |
|---|---|---|
| parameters | 0.75B | 0.86B (ViT 100M + point head 9M) |
| question types | choice / score / noul | choice / score / noul / point |
| state | text or JSON | text or JSON + up to 4 images |
| typical use | routing, moderation, RAG relevance, tool choice, agents with element tables | GUI agents from pixels, Thai voice control, Thai screen reading, game frames, documents |
| text-only behaviour | v0.3 | same API; text scores equal or better than v0.3 (tables below) |
/v1/systemone contract){
"state": "เป้าหมาย: ค้นหาเที่ยวบิน BKK→CNX",
"images": [{"id": "screen", "data": "data:image/png;base64,...", "role": "screenshot"}],
"questions": {
"where": {"type": "point", "instructions": "ปุ่มค้นหาเที่ยวบิน", "image": "screen"},
"op": {"type": "choice", "instructions": "next operation", "criteria": {"CLICK": null, "TYPE_TEXT": null, "SCROLL_DOWN": null, "DONE": null}},
"loaded": {"type": "noul", "instructions": "is the page fully loaded?"}
}
}
point answers: {"x": 0.63, "y": 0.41, "confidence": 0.86, "abstain": 0.02, "candidates": [{"x", "y", "p", "bbox"}, ...]}
(normalised image coordinates; abstain = probability that the target is not on the image; candidates = the
above-threshold token clusters of the attention heat-map, GUI-Actor style).
Limits: ≤ 4 images, ≤ 8k visual tokens per request (a 1280×800 screenshot ≈ 1,000 tokens at 32 px/token; a 320×240 game
frame ≈ 80); "detail": "high" allows up to 1920×1200.
Code status (2026-11-15). The Python client shown on this card (image inputs,
point,point_refine) is not in the public GitHub repository yet; it arrives with the next code update there. What runs today with public code: the GGUF build iapp/OpenThai-SystemOne-Vision-GGUF on stock llama.cpp (llama-server,/v1/systemone): choice / score / noul questions, with or without a screenshot. It has nopointhead, so a click there is a choice among the controls the app reads from the screen (see that card).
from openthai_systemone import SystemOneClient
c = SystemOneClient("iapp/OpenThai-SystemOne-Vision")
r = c.system_one("goal: open the login form", {"where": {"type": "point", "instructions": "ปุ่มเข้าสู่ระบบ", "image": "s"}},
images=[{"id": "s", "path": "screen.png", "role": "screenshot"}])
print(r.answers["where"].x, r.answers["where"].y, r.answers["where"].confidence)
An app that can read the controls of the front window through the accessibility API (UI Automation, the macOS Accessibility API, AT-SPI) does not need a screenshot to click: it lists the controls and asks which one the command means.
r = c.system_one("os: windows\napp: CalculatorApp.exe\ngoal: กดปุ่มบวก",
{"target": {"type": "choice", "instructions": "Which element should be acted on to achieve the goal?",
"criteria": {"0": 'button "Seven"', "1": 'button "Plus"', "2": 'button "Equals"', "none": "none of these"}}})
print(r.answers["target"].choice, r.answers["target"].probabilities) # "1"
os: windows | macos | linux, app: <program>, goal: <the command>, one per line;button "Save", menuitem "View", listitem "General",
the Windows type words on every OS), in reading order, up to about 120 per question, "none": "none of these" last;none when nothing listed fits. v1.5 says so for 51 to 78 % of the commands whose control is missing (v1.4: 8
to 15 %), so keep a check in the app before it clicks, for example that the chosen control's name is related to the command.The voice-command schema (21 operations such as SEARCH, CLICK_NTH, OPEN_WEBSITE, QUIT_APP, PLAY_PAUSE, LIKE,
COMMENT, NONE = not a command, plus search_site / app / site / key / nth slots) and a macOS controller
(on-device Thai speech recognition → this model → native mouse / keyboard events) will be published in the GitHub
repo under demo/voice_control/. One text-only call decides the operation; clicks add one screenshot call.
state = 'คำสั่งเสียงภาษาไทยจากผู้ใช้ที่ควบคุมคอมพิวเตอร์ด้วยเสียง: "อยากดูเนเน่ร้องเพลง"\nโปรแกรมที่ใช้อยู่: Google Chrome\nหน้าต่าง: Facebook'
# questions = intent_schema.intent_questions(apps) -> operation SEARCH, search_site YouTube
v1.5 was trained and measured on a second machine. v1.4 was measured again there with the same scripts, so every pair below compares like with like; the v1.4 figures differ a little from the older tables further down, which come from the first machine. Accuracy in %; "default" and "+ refine" as defined under Earlier versions.
Element choice in desktop apps (the question under "Clicking in a desktop app"; text only unless noted; the test controls' names are never asked in training, and the natural sentences were written by hand):
| the command | v1.4 | v1.5 |
|---|---|---|
| Windows 11 (28 screens of 9 apps in 7 looks, about 1,650 targets, lists of up to 120 controls) | ||
| · names the control as written on screen | 99.0 | 99.6 |
· a Thai verb around the label as written (กดปุ่ม Save) |
99.3 | 99.5 |
| · a natural English sentence | 96.0 | 98.0 |
| · a natural Thai sentence, English labels on screen | 86.1 | 93.2 |
| · an English command, Thai labels on screen (72 targets, one app) | 94.4 | 95.8 |
| · natural Thai / English with the screenshot attached | 81.6 / 96.6 | 82.3 / 97.7 |
· control not listed: answers none (natural Thai / natural English / name as written) |
8 / 9 / 15 | 51 / 67 / 75 |
Ubuntu 24.04, GNOME apps (os: linux, 178 targets, 160 with a natural sentence) |
||
| · names the control as written / a Thai verb around it | 98.9 / 99.4 | 99.4 / 98.3 |
| · a natural Thai sentence | 78.8 | 89.4 |
| · a natural English sentence | 89.4 | 90.6 |
· control not listed: answers none (natural Thai / natural English / name as written) |
10 / 11 / 11 | 58 / 79 / 84 |
On macOS control lists v1.4 chose the control named in the command 99.3 % of the time (560 targets); v1.5 was trained on those screens, so it has no held-out macOS figure yet.
Pointing on live desktops (point; click inside the control read from the accessibility tree; screens never trained on):
| benchmark | v1.4 default | v1.4 + refine | v1.5 default | v1.5 + refine |
|---|---|---|---|---|
| Windows 11 in a VM, 7 looks: light / dark, accent colours, taskbar left, 100 to 175 % scale, window and full screen (1,754 targets) | 72.9 | 77.0 | 76.2 | 80.0 |
| Windows 11 laptop at its own scales: 2880×1800 at 200 and 175 %, 1920×1080 and 2560×1600 at 150 %; some apps Thai-labelled (839 targets) | 66.0 | 72.2 | 70.7 | 77.0 |
| Windows live, the set of the tables below (318 targets) | 71.1 | – | 72.6 | – |
| Ubuntu 24.04 GNOME, 6 looks: default, light, dark, high contrast, large text, 200 % (1,260 targets) | 72.5 | 76.5 | 76.8 | 79.2 |
| ScreenSpot-v2, all (1,272) | 81.9 | 84.3 | 81.5 | 83.3 |
| · desktop text / icon | 88.7 / 75.7 | 90.2 / 82.9 | 88.7 / 72.1 | 89.2 / 79.3 |
| · mobile text / icon | 94.8 / 71.1 | 94.5 / 73.5 | 94.8 / 73.5 | 93.1 / 78.7 |
| · web text / icon | 85.5 / 68.5 | 87.2 / 72.9 | 84.2 / 67.5 | 85.5 / 69.0 |
| Thai GUI held-out domains (600) | 75.0 | 75.0 | 75.4 | 75.4 |
| OS-Atlas desktop, held-out screens (1,028: macOS 137, Windows 680, Linux 211) | 52.7 | 58.1 | 52.4 | 58.1 |
| · macOS / Windows / Linux, + refine | – | 70.8 / 56.2 / 55.9 | – | 70.1 / 56.5 / 55.5 |
Known regression: on public screenshots v1.5 is behind v1.4, most of all with zoom: ScreenSpot-v2's Windows shots 89.9 →
87.4, its macOS shots 86.4 → 85.0, its desktop icons 82.9 → 79.3, its web icons 72.9 → 69.0 (all of ScreenSpot-v2: 84.3 →
83.3). OS-Atlas desktop screens are level (macOS 70.8 → 70.1, Windows 56.2 → 56.5, Linux 55.9 → 55.5): v1.5's last round
rehearsed that public set again, on other screens than the ones tested. The Jedi icon set v1.4 had is still missing. Not
measured again for v1.5: Mac live, ScreenSpot-Pro. If you point on macOS, on web pages or in professional desktop apps,
compare with v1.4 (revision="v1.4") before switching.
Voice commands, Doom, text:
| set | v1.4 | v1.5 |
|---|---|---|
| real voice commands spoken in use, operation (183) | 90.7 | 89.1 |
| · slots when the operation is right | 100.0 | 100.0 |
| · chit-chat between commands acted on (41), lower is better | 7.3 | 4.9 |
| real off-command conversation acted on (227), lower is better | 4.0 | 1.8 |
| "click" + a name on the screen, held-out names (292: must be a click, a key or the n-th item) | 78.1 | 95.2 |
| · of these, key commands such as "กด Enter" (24) | 95.8 | 95.8 |
| Doom from pixels, kills per episode (seed 999, 40 episodes) | 10.4 | 10.6 |
| text: public 13-subset macro / Thai 8-set macro | 74.9 / 79.2 | 75.3 / 79.5 |
On the real commands v1.5 is 3 of 183 behind v1.4: it reads a few indirect requests as not a command. It acts on background talk less often and reads "click + a name" far better.
v1.4 next to v1.3, same scripts; single option order unless noted; accuracy in %. "default" = the plain API call (screenshot at
≤ 1280×800); "+ refine" = screenshot at 2 MP ("detail": "high") with point_refine over the top-3 first-pass candidates,
landscape screens only (zoom 50 % on ScreenSpot-v2 / Thai GUI / OS-Atlas, 35 % on Windows / Mac live and ScreenSpot-Pro).
On desktops and large screens, send "detail": "high" and point_refine.
| benchmark | v1.3 default | v1.3 + refine | v1.4 default | v1.4 + refine | Qwen3-VL-2B | Qwen3-VL-4B |
|---|---|---|---|---|---|---|
| Windows live (319 targets, 28 screens of Windows 11: Settings, File Explorer, Control Panel, Notepad, Paint, Calculator, Store, Edge, …) | 68.3 | 69.3 | 70.8 | 70.5 | 67.1 | 73.0 |
| · Windows Settings pages (114) | 69.3 | – | 74.6 | – | 66.7 | 68.4 |
| · Windows Calculator (20)† | 60 | – | 80 | – | 80 | 95 |
| Mac live (358 targets, 29 screens: System Settings, Finder, Safari, Office, iWork, …) | 63.7 | 67.6 | 64.5 | 67.3 | 67.0 | 67.9 |
| · label-checked (340: 18 targets found unanswerable removed) | 66.8 | 71.2 | 67.6 | 70.9 | 70.6 | 71.5 |
| latency per click (H100, eval harness) | ~70 ms | 0.1–0.25 s | ~70 ms | 0.1–0.25 s | ~0.9 s | ~0.75 s |
The Windows and Mac live screens are never trained on; the training data includes other screens of the same apps. † v1.4's synthetic calculators use Windows' standard accessible names ("Two", "Multiply by"), which is how the Windows live Calculator targets are named, so part of that gain is learning the naming convention.
point: predicted click inside the target box)| benchmark | v1.3 default | v1.3 + refine | v1.4 default | v1.4 + refine | Qwen3-VL-2B | Qwen3-VL-4B |
|---|---|---|---|---|---|---|
| ScreenSpot-v2, all (1,272) | 82.2 | 85.2 | 82.8 | 85.9 | 85.8 | 90.4 |
| · desktop text / icon | 90.2 / 75.7 | 91.8 / 83.6 | 89.2 / 77.1 | 91.2 / 85.0 | 93.8 / 82.9 | 96.9 / 85.7 |
| · mobile text / icon | 94.5 / 71.6 | 93.8 / 74.9 | 94.8 / 71.6 | 94.5 / 75.4 | 94.1 / 82.0 | 96.9 / 83.4 |
| · web text / icon | 84.2 / 70.0 | 88.5 / 74.9 | 85.9 / 71.4 | 89.3 / 76.4 | 85.9 / 72.4 | 91.9 / 83.7 |
| ScreenSpot-Pro (1,581, 26 pro apps up to 4K; Windows / macOS / Linux) | 9.0 | 32.9 (29.3 / 38.7 / 28.0) | 8.7 | 32.8 (29.2 / 38.9 / 26.0) | 43.4* | 54.8* |
| Desktop (OS-Atlas macOS / Windows / Linux, held-out) | 53.7 | 59.8 (600) | 53.8 | 59.3 (600) | 63.8 | 65.6 |
| Desktop "not on screen" abstention (26) | 92.3 | – | 92.3 | – | – | – |
Thai GUI held-out domains, point (1,611) |
77.8 | 76.7 (600) | 78.6 | 74.9 (600) | 72.4 | 80.4 |
Thai GUI set-of-marks choice / operation |
93.1 / 99.6 | – | 92.6 / 99.8 | – | – | – |
Thai screen QA noul / choice |
93.1 / 85.9 | – | 94.7 / 85.0 | – | – | – |
| Thai GUI "not on screen" abstention (179) | 48.0 | – | 49.7 | – | – | – |
Qwen3-VL numbers: our runs with the same harness (screenshot + instruction, click inside the box); * every 2nd ScreenSpot-Pro item (790). Mac live and the 600-item Thai GUI / OS-Atlas subsets move by ±1.5 points between settings; smaller differences are noise.
point_refiner = c.system_one("goal: …", {"where": {"type": "point", "instructions": "ปุ่มบันทึก", "image": "s"}},
images=[{"id": "s", "path": "screen.png", "role": "screenshot", "detail": "high"}],
point_refine={"zoom": 0.35, "cands": 3}) # HTTP API: "point_refine": {...} or true
The first pass points on the whole screen; the client then re-points on a crop around each of the top candidates at native
detail and keeps the most confident. It only zooms landscape screens larger than the pixel budget (zooming phone screenshots
hurts). v1.4: ScreenSpot-Pro 8.7 → 32.8, ScreenSpot-v2 82.8 → 85.9, Mac live 67.6 → 70.9 (checked). Windows windows usually fit
the 2 MP budget, so there it mostly adds the higher resolution. Cost: 1 + cands forward passes.
| set | v1.1 | v1.3 | v1.4 |
|---|---|---|---|
| real commands spoken in use, operation (183) | 88.5 | 89.6 | 90.7 |
| · slots (site / app / key / Nth) when the operation is right | 100.0 | 100.0 | 100.0 |
| · chit-chat between commands acted on (41), lower is better | 7.3 | 7.3 | 7.3 |
| real off-command conversation picked up by the microphone, acted on (227), lower is better | 2.6 | 4.4 | 4.0 |
| synthetic voice commands, held out (1,406 questions) | 90.5 | 92.0 | 91.8 |
| synthetic hard cases (fragments, site vs app, ordinals, 446 questions) | 90.8 | 93.3 | 92.6 |
demo/doom_demo.py, defend_the_center, skill 3, two 320×240 frames, 40 episodes per seed)| seed | v1.1 | v1.3 | v1.4 | ms / decision |
|---|---|---|---|---|
| 999 | 11.8 | 13.0 | 11.4 | ~50 |
| 777 | – | 11.9 | 9.7 | ~50 |
| mean of both (80 episodes) | – | 12.5 | 10.6 | |
| the privileged scripted expert it imitates | 6.8 | 6.8 | 6.8 | – |
Known regression: v1.4 plays Doom from pixels worse than v1.3 (−1.9 kills per episode). Use v1.3 (revision="v1.3") for the
Doom demo. Offline Doom decisions (600 frames), v1.4: move 88.5, attack 97.2, in-danger 96.7, nearest-enemy point 48.9,
enemy-absent 98.3.
Public 13-subset benchmark (the Bespoke / Jev public suite; noul/choice = accuracy, score = exact level):
| subset | v0.3 | v1.3 | v1.4 | Jev 1.13.0* | Nimble-9B* |
|---|---|---|---|---|---|
| aegis2 | 83.2 | 82.0 | 81.2 | 80.4 | 81.2 |
| boolq | 79.7 | 79.7 | 79.0 | 89.7 | 86.0 |
| civil_comments | 79.0 | 86.3 | 86.3 | 81.0 | 70.3 |
| helpsteer2 (score) | 41.6 | 40.4 | 40.4 | 34.1 | 39.0 |
| massive-de-DE | 88.3 | 89.7 | 90.0 | 86.9 | 83.4 |
| massive-en-US | 88.3 | 90.3 | 90.3 | 87.4 | 86.9 |
| multinli | 89.0 | 88.3 | 89.0 | 82.9 | 85.3 |
| paws | 94.0 | 94.0 | 94.0 | 89.2 | 82.8 |
| pubmedqa | 64.0 | 67.2 | 67.2 | 77.2 | 75.6 |
| squad2 | 89.3 | 88.6 | 88.3 | 82.9 | 80.6 |
| summeval-consistency (score) | 75.0 | 75.0 | 75.7 | 81.2 | 75.7 |
| summeval-relevance (score) | 21.7 | 20.4 | 20.4 | 35.0 | 49.2 |
| vitaminc-dev | 72.5 | 71.0 | 70.8 | 80.1 | 76.6 |
| macro, 13 subsets | 74.3 | 74.8 | 74.8 | 76.0 | 74.8 |
| macro, 10 accuracy subsets | 82.7 | 83.7 | 83.6 | 83.8 | 80.9 |
* published by Bespoke Labs for their own harness on the same subsets (Jev via the TypeSafe API).
Thai held-out and eval-split sets (th unless noted; banking77 and xLAM are English):
| set | v0.3 | v1.3 | v1.4 | v1.4, 8 orders |
|---|---|---|---|---|
| banking77 (77-way, whole dataset held out) | 45.4 | 62.0 | 62.1 | 64.4 |
| contrastive pairs, one-fact flips | 80.7 | 82.1 | 81.8 | 80.1 |
| MASSIVE-th intent | 90.0 | 91.1 | 91.2 | 92.0 |
| Prachathai topics | 98.1 | 98.6 | 98.6 | 98.7 |
| SIB-200 Thai topic (whole dataset held out) | 77.9 | 75.5 | 75.0 | 77.0 |
| Wisesight sentiment (whole dataset held out) | 51.6 | 52.5 | 52.2 | 51.7 |
| Wongnai stars (score, exact) | 63.5 | 63.3 | 62.9 | 62.9 |
| xLAM tool selection | 99.4 | 99.4 | 99.4 | 99.3 |
| XNLI-th | 77.1 | 75.5 | 75.6 | 75.7 |
| macro, 8 accuracy sets | 77.5 | 79.6 | 79.5 | 79.9 |
SystemOneClient)| request | tokens | ms |
|---|---|---|
| two 320×240 game frames + short state, 2 questions | 298 (160 visual) | 50 |
1280×800 screenshot + goal, point + choice |
1,062 (1,000 visual) | 52 |
large screenshot at "detail": "high" + point_refine (3 candidates), eval harness |
~2,000 visual per pass | ~135 (FHD) – 245 (4K) |
| text only (3-question Thai ticket) | 166 | ~40 |
| version | date | change |
|---|---|---|
| v1.5 | 2026-11-20 | + element choice for desktop apps (41k text-only questions over the control lists of live Windows 11 screens, 3.6k over macOS and Ubuntu lists; natural Thai and English wordings, Thai-labelled copies, 8 to 30 % with the control left out; 62k Thai ↔ English interface-vocabulary questions built from the Thai translation catalogs of open-source desktop software), pointing on 5,350 new live Windows 11 screens in 24+ looks (a VM and a laptop; 100 to 300 % scale, light / dark, accent, full screen) and on Ubuntu GNOME, zoomed crops, voice "click + a name" records, public desktop screenshots (OS-Atlas, 24k) and their zoomed crops rehearsed again; six rounds from v1.4. Windows natural-Thai choice 86.1 → 93.2, Ubuntu 78.8 → 89.4, none for a missing control 8–15 → 51–78 %, pointing on live Windows screens + 2 to 5, Ubuntu + 3 to 4, voice click-named 78.1 → 95.2, text unchanged; ScreenSpot-v2 with zoom: desktop screenshots − 1 to 4, web icons − 4; real voice commands 90.7 → 89.1. |
| v1.4 | 2026-09-28 | + icon grounding and "not on screen" refusals from xlangai/Jedi (20k icon snaps, 20k refusal rows), 12k synthetic calculators / title bars / toolbars / media controls labelled with their standard accessible names (English + Thai), Mac and Thai GUI rehearsal up; from v1.3. Windows live 68.3 → 70.8 (Settings 69.3 → 74.6, Calculator 60 → 80), ScreenSpot-v2 82.2 → 82.8 (85.9 with refine), Thai not-on-screen 48.0 → 49.7, voice 89.6 → 90.7; Doom from pixels 12.5 → 10.6 kills; text, Thai, Mac live, ScreenSpot-Pro, OS-Atlas unchanged within noise. |
| v1.3 | 2026-09-26 | + live desktop grounding: 665 records from 319 live Windows 11 screens (UI Automation targets; Settings, Control Panel, dialogs, shell folders, Calculator modes, Store, Edge) and 386 records from 116 live macOS screens (accessibility targets; System Settings, Finder, utilities, Safari, Word / Excel / PowerPoint / Pages ribbons), with account names blacked out; on top of an internal v1.2 round (ShowUI desktop + web, 40k zoomed crops, 75k xlangai/Jedi records, 2 MP training, averaged with v1.1). Client / API point_refine. Windows live 63.3 → 68.3, ScreenSpot-v2 80.7 → 82.2 (85.2 with refine), ScreenSpot-Pro ~10 → 32.9 with refine, OS-Atlas desktop 50.6 → 53.7, Doom 11.8 → 13.0, text and Thai unchanged. |
| v1.1 | 2026-09-24 | + Thai voice-command intents (28k synthetic training records from 21k distinct commands + 8.9k hard-case records: fragments / chit-chat = NONE, site vs app, ordinal + topic, ASR spelling variants), desktop grounding round (OS-Atlas macOS / Windows / Linux), text-anchor round (civil_comments, VitaminC); temperatures recalibrated. Voice 65 → 88.5, desktop 39 → 51, Doom pixels 5.5 → 11.9 kills, ScreenSpot-v2 82.7 → 80.7. |
| v1.0 | 2026-09-23 | first vision release (Thai GUI crawl, public grounding, Doom rollouts, screen QA on the v0.3 text tower) |
move / attack / in_danger / nearest_enemy
questions, and screen QA.point_refine second pass; 75k xlangai/Jedi records), with voice, text and Doom rehearsed; the release is
the equal-weight average of v1.1 and the last round, which kept v1.1's Thai GUI and Doom skills and the new desktop gains.point_refine second pass, and element-choice questions without a screenshot. The commands are the label as
written, template sentences, and natural Thai / English sentences written by a teacher LLM from a hand-checked Thai
glossary and the official English / Thai names of the same controls, each checked blind against the look-alike controls of
its screen; a share of them never repeats the label. Thai ↔ English interface vocabulary from the Thai translation catalogs
of open-source desktop software (the GNOME and Ubuntu language packs, LibreOffice): text-only element-choice questions that
ask for a control by the translators' Thai term among the English labels of the same program, and the reverse; labels that
are test targets are left out. Voice-intent records "click verb + a real control name" with key and ordinal contrasts.
Text, Thai GUI, public grounding and symbol screens rehearsed. Public desktop screenshots (OS-Atlas) and zoomed crops of
them rehearsed in the last round. The captured screens are not published.Apache-2.0, built on Qwen3.5-0.8B-Base. Developed by iApp, launched with AIEAT. GUI grounding recipe after GUI-Actor; System One contract after TypeSafe AI's Jev. Grounding data from OS-Atlas, ShowUI, Wave-UI, GroundUI and Jedi (OSWorld-G); benchmarks ScreenSpot-v2 and ScreenSpot-Pro.