Amos

Modelsamos-1-large

Amos 1 largeOur most capable model, for the hardest calls.

Try it in the playground
  1. Size27B
  2. Server time, median122 ms
  3. Context128K tokens
  4. Typed decisions, zero-shot76.1%
  5. Typed decisions, fitted80.3%
  6. Calibration, KL0.126
  7. Decision Bench94.9%
  8. Decision Index 0.3, public65.11

When to use it

  • Questions that take several steps.
  • Long policies and trade-offs.
  • Questions that need broad knowledge.
  • The best Decision Index, photo and calibration scores of the four.
  • Cases up to 128K tokens.

Typed decisions

Accuracy on the 2,000-decision test split, zero-shot. We ran each rival on the same 2,000 decisions; Decider 1's score is its makers' own. Frontier language models asked the same questions score between 73 and 78.

  1. Microsoft-Decision-1closed weights, KL 0.31879.5
  2. Decider 176.8
  3. Amos 1 medium76.5
  4. Amos 1 large76.1
  5. Perplexity Decider v1.173.9
  6. Clef72.5
  7. GPT-6 Luna71.9
  8. Amos 1 small69.5
  9. Amos 1 mini64.6

Quotes and photos

  1. Invoice totals quoted exactly10 / 10
  2. Box centred on the named object, RefCOCO91.9%
  3. Boxes covering at least half the object, RefCOCO71.3%
  4. Box centred on the named object, COCO98.4%

Limits

  • The slowest of the four.
  • Long quotes, such as a whole clause or address, can come back cut short.
  • English only.

Quick start

Name the model in the model field.

curl https://api.amos.scira.ai/v1/systemone \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "amos-1-large",
  "state": "Hi, invoice INV-2231 for £306.00 was charged to my card twice this morning. Please refund one of them today.",
  "questions": {
    "refund": {
      "type": "noul",
      "instructions": "Is the customer asking for a refund?"
    },
    "team": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": {
        "billing": null,
        "technical": null,
        "sales": null
      }
    },
    "amount": {
      "type": "span",
      "instructions": "How much was charged?"
    }
  }
}'

The other models