emrevrg/AUBIN-Omni

Open, calibrated decision models that see, act and learn.

Typed decisions · computer use · real-time control · self-learning, open weights on Gemma 4.

Omni · 31B · 12B · E4B · Screen · Web · Control · Support ❤

♥ If AUBIN impresses you, press Like at the top of this page. It is the single biggest help for an independent open model.

_{🎬 AUBIN in 80 seconds (sound on) · Türkçe izle · vertical reel: EN / TR
Music: “Aphelion” by Scott Buckley, released under CC BY 4.0 · scottbuckley.com.au}

AUBIN Omni: one model, every ability

AUBIN Omni loads the Gemma-4 base once. Each ability is an AUBIN adapter that plugs in on top and switches in milliseconds. An adapter is loaded lazily, the first time its ability is used, so start-up only pays for the base model and one ability.

ability	call	E4B (fast)	12B	31B
typed, calibrated decisions	omni.decide(state, questions)	✅	✅	✅
click on screenshots	omni.click(image, "open settings")	✅ ScreenSpot 69.3	✅ ScreenSpot 66.7 ¹	✅ via E4B-Screen ²
web agent step	omni.web_step(task, candidates)	✅	✅ Mind2Web 43.5	✅ Mind2Web 45.0 ³
real-time control and games	omni.act(obs, command, actions, allowed=…)	✅	✅ 92% · 0 lava	✅ via E4B-Control ²
self-learning and new skills	omni.enable_learning() → learn · acquire_skill	✅	✅	✅

¹ The Gemma-4-12B base itself (no AUBIN adapter), run with fp32 compute: in fp16, image tokens overflow the 12B language model and every answer comes out empty. Full ScreenSpot, 1,272 screenshots. AUBIN fine-tuning of the 12B screen ability is the next training run.
² Companion model: this ability runs on a separate, lazily loaded E4B adapter (measured on its own card), so one omni object still offers every ability. omni.companions lists them.
³ AUBIN-31B web adapter, Mind2Web MindAct protocol (cross-domain split, 200 steps).

The ability manifest lives in omni.json. New abilities appear here as they are trained, and the code does not change.

Use

pip install "git+https://huggingface.co/emrevrg/AUBIN-12B#subdirectory=code"

from aubin import AubinOmni
omni = AubinOmni.from_hub("emrevrg/AUBIN-Omni", size="E4B")    # or "12B", "31B"
print(omni.abilities())                                          # ['control', 'decide', 'screen', 'web']

omni.decide("Billed twice for March, refund or we cancel.",
            {"team": {"type": "choice", "instructions": "Which team?", "criteria": {"billing": "payments", "tech": "bugs"}},
             "churn": {"type": "noul", "instructions": "Does the user threaten to cancel?"}})
omni.click(screenshot, "open settings")                          # {"x": 214, "y": 731} on a 0-1000 scale
omni.web_step("Book a flight from New York to Paris", {"n1": "<input placeholder='From'>", "n2": "<a>Hotels</a>"})
omni.act("lava above, key up-right", "go to the red key",
         {"up": "row-1", "down": "row+1", "left": "col-1", "right": "col+1"}, allowed={"left", "right", "down"})
smart = omni.enable_learning()                                   # learns from feedback; adds skills only after a self-test

Measured smoke test

Setup: one free T4, Gemma-4-E4B in 4-bit, all four adapters on one model. Code: code/omni_smoke.py.

call	result
decide	"billing" at 0.91, "will cancel" yes at 0.87
web_step	the "From" input with TYPE, at 0.99 / 0.98
click	70% correct on 30 ScreenSpot samples, about 2.6 s per click
act	the model's move into lava was overridden by the safety shield, which picked the safe "left"
switch back to decide	"technical" at 0.90

The 30-sample click test is only a smoke check. The full ScreenSpot score (1,272 screenshots) is 69.3 for AUBIN-E4B-Screen.

---

_{🎬 AUBIN filmi, Türkçe (80 sn, sesi aç) · English · Müzik: “Aphelion”, Scott Buckley, CC BY 4.0}

♥ Help AUBIN get seen

On Hugging Face, likes decide what people discover. Big labs have marketing teams; AUBIN has one 17-year-old student and measured results. If AUBIN is useful, interesting or just impressive to you, press ♥ Like at the top of this page and on AUBIN-Omni, AUBIN-31B and AUBIN-12B, then share it with one person who builds with AI. Every like helps an independent, open, honestly measured model get discovered.

_{🇹🇷 Beğenin, AUBIN'in görünür olmasını sağlar: sayfanın üstündeki ♥ Like'a basarak destek ol ve bir arkadaşına gönder.}

Support Norovox

Built by a 17-year-old high school student: no sponsor, no budget, just free GPUs and AI subscriptions paid for with difficulty. Support goes into GPU compute, training, and the AI development tools this work depends on (such as Claude); supporters are credited and get early access. zgremre@gmail.com · emrevrgdev@gmail.com · Why and how →

_{🇹🇷 17 yaşında bir lise öğrencisinin eseri: destekçisiz, bütçesiz; ücretsiz GPU'lar ve zorlukla ödenen yapay zekâ abonelikleriyle. Desteğin GPU'ya, eğitime ve bu işin dayandığı yapay zekâ geliştirme araçlarına (Claude gibi) gider.}

_{Built with Claude Opus 5.5 and GPT-5.6 Sol; because of OpenAI usage limits, the final stretch was completed with Claude Opus 5.5. License: Apache-2.0 (adapters and code). Base models: Google Gemma 4 (Apache-2.0).}