NewNimble models: #1 decision models; outperforming Jev 1.13

Nimble · Decision models · Hosted API

Decisions,
with the [odds] attached.

Nimble turns text into decisions. Send the state and the answers you allow, and it returns a probability for every option, so your code can act on the confident calls and hand the close ones to a person.

#1 on the Decision Index, ahead of Jev 1.13.

Top public score · any size

68.46†Nimble-v4-max 27B

+10.50 vs Jev 1.13

#1 of 104 models under 25B

63.70†Nimble-v4 9B

+5.74 vs Jev 1.13

Both models, one API, one request format.

† Public-suite score from our own run, Oct 12, 2026. Not yet on the board.

Chess bench

Nimble vs. Jev, move by move.

Leaderboard ranking · Decision Index · Public suite

State of the art at 9B and 27B.

Pushing the Pareto frontier. #1 on Decision Index public suite.

  • Nimble-v4 63.70†
  • Nimble-v4-max 68.46†
  • Perplexity Decider v1.1 62.25
  • Board models (143)
  • Pareto frontier

Served parameters (log scale)

† Public-suite score from our own run, Oct 12, 2026. Not yet on the board.
Others: Decision Index 0.3.2 board data (huggingface.co/spaces/multimodalart/jev-decision-index), Oct 12, 2026. The public suite is 20% of the board's full score.

How it works

One request, a probability for every answer

Send a state and your questions in one request. Nimble reads the state once and returns an answer for every question, with the probability it puts on each option. Both models take the same request.

State

Hi, I was charged twice for my March invoice (#4471). Can you refund one of them?

Question

Which team should handle this ticket? billing · technical · account · sales

Nimble's answer
billingConfident: send it straight to the billing queue.
billing94%
account3%
technical2%
sales1%

Example output for illustration.

curl https://api.bespokelabs.ai/v1/systemone \
  -H "Authorization: Bearer $BESPOKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nimble-latest",
    "state": "The customer was charged twice and asks for a refund. The service still works.",
    "questions": {
      "refund": {
        "type": "noul",
        "instructions": "Does the customer request a refund?"
      },
      "department": {
        "type": "choice",
        "instructions": "Which department should handle this request?",
        "criteria": {
          "billing": "Charges, payments and refunds",
          "technical": "Software bugs and outages"
        }
      }
    }
  }'

# → {"answers": {"refund": {"type": "noul", "noul": 0.998},
#      "department": {"choice": "billing", ...}}}

From the Nimble quickstart. The response is shortened.

noul

Is this true?

Ask whether a statement is true. The answer is the probability that it is, from 0 to 1.

choice

Which option fits best?

Name the options. You get the chosen one and a probability for each.

score

Where does it fall?

List the levels from lowest to highest. You get the expected level and a probability for each one.

Decisions, with the odds attached.

Sign up