Aplomb 1 Omni

Aplomb 1 Omni answers typed questions about text, JSON, images, video, and audio with calibrated probabilities. You send a state and a set of questions; every question comes back as a probability distribution with a confidence value, in one request, and no text is generated.

• Question types: noul (yes or no), choice (one of 2 to 255 options), score (2 to 10 ordered levels), and tool (which function to call from a list of function tools, with distributions over its enum and boolean arguments).
• Inputs: text, JSON objects and arrays, images, video, and audio (speech, sounds, and music). For text, images, and video only, Aplomb 1 scores higher on most benchmarks below.
• Independent answers: questions never influence each other.

Results

Measured on 2026-09-27 with the weights in this repository, served the way the hosted model is. Accuracy is in percent; for calibration, lower is better. Aplomb 1 Omni was not trained on any benchmark below unless it is marked (a), and inputs that also appear in its training data were removed from scoring for every system. Other models were run with their own released inference code.

Audio

Decision benchmarks (the Intern-Decision evaluation bundle)

Calibration (lower is better)

Text classification ((k): Kev-27B was trained on this dataset or its source)

Images

Video

(a) Another split of the same dataset was used in training. (c) Jev (jev-1.13.0) figures as published by Intern-Decision for the same items. (b) 5 of its 19 subtasks use CLEVRER, whose training split was used in training.

Quickstart

curl https://api.empiriolabs.ai/v1/decisions \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "aplomb-1-omni",
    "state": {
      "call": {"type": "audio", "url": "https://example.com/support-call.wav"},
      "account_tier": "business"
    },
    "questions": {
      "wants_refund": {"type": "noul", "instructions": "Does the caller ask for a refund?"},
      "tone": {"type": "score", "instructions": "How upset does the caller sound?",
               "criteria": ["Calm", "Annoyed", "Angry"]}
    }
  }'

Every answer comes back with its probabilities and a confidence value. The full request and response schema, media input, limits, and pricing are in the Decisions API guide and on the model page.

Run locally

pip install -U "transformers>=5.17" torch accelerate torchvision torchcodec
python run_aplomb.py --model empiriolabsai/aplomb-1-omni --request request.json --debias

request.json is a Decisions API request body; --debias matches the hosted API. The weights are bf16 and need about 70 GB of accelerator memory.

Limits

• The state plus the longest question: up to 60,000 tokens.
• Media: up to 16 images, 8 audio clips (10 minutes each), and 4 videos per request.
• Aplomb 1 Omni does not generate text and does not fill free-text tool arguments; those are listed under open_arguments.

License

EmpirioLabs Model License 1.0. Research, education, evaluation, personal use, and internal commercial use by organizations with annual revenue under US$1 million are free. Offering the model as a hosted service, commercial use above that threshold, and distributing it in products sold to others need a commercial license: write to support@empiriolabs.ai. Using the model or its outputs to distill or train another model is not permitted. The base model is licensed under Apache 2.0; its license text and the attribution notice are in LICENSE-APACHE-2.0 and NOTICE.