What Cicero is
Cicero is a decision model from Butler Compute. You send a state (text, a chat, or up to four images) and up to 64 typed questions. It returns a calibrated probability for every option of every question and writes no text. It takes the three question types Jev uses: Noul for yes or no, Choice for one of up to 255 options, and Score for a rating on a scale.
Butler built it as a post-train of Qwen3.5-35B-A3B, a mixture-of-experts model with 35B parameters, of which 3B are active for each token. The vision tower comes from the base model unchanged.
How it answers
Butler’s card says each question is laid out as plain text: the context, then the question, then its options lettered A, B, C. The model reads that prompt in one forward pass. Butler reads the answer from the model’s scores for the option letters at the end of the prompt, turns them into probabilities, and calibrates them for each question type.
What Butler reports
Every quality figure here is vendor-run. Butler ran the public suite of the Jev Decision Index 0.4.1 with the official kit and reports a public score of 71.33. Its best area is Tools & Automation at 87.0 and its weakest Knowledge & Reasoning at 56.1. Cicero had no row on the board itself as of 8 November 2026. The board’s headline Full score counts only private tests, so it has no comparable figure for Cicero yet. Butler gives a median latency of 41 ms without saying how it measured it.
What it’s good at
Cicero suits high-volume typed calls where cost matters: routing a support ticket, picking an intent from a long list, or deciding whether an agent should call a tool. The card’s own examples are a 77-way banking intent question and a set of yes or no tool-call questions. It also reads images and long contexts.
What it’s not for
It does not generate text. Until an independent board scores it, treat the quality numbers as Butler’s claims.
Access today
Cicero is live on Butler’s API at api.butlercompute.com, with a System One endpoint and an OpenAI Decisions endpoint, both behind a bearer key. Butler lists it at $0.01 per million input tokens with output free, level with Span-02 as the lowest input price of any commercial model on this site as of 8 November 2026. It is not on OpenRouter. The weights are on Hugging Face under Apache 2.0, with a smaller NVFP4 quantized copy.
Specifications
| Question types | ChoiceScoreNoul |
| Max Choice options | 255 |
| Score levels | Not documented |
| Questions per call | 64 |
| Total context | 131,072 tokens |
| State budget | Not documented |
| Rate limit | Not published as of 2026-11-08. |
| Endpoint | POST https://api.butlercompute.com/v1/systemone (System One format) or POST https://api.butlercompute.com/v1/decisions (OpenAI Decisions format), with a bearer key |
| Licence | Apache-2.0 |
| SDKs |
The model card gives a 131,072-token context and inputs of text, a chat, or up to 4 images, with up to 64 questions per request and up to 255 options per Choice question. It publishes no cap on Score levels.
Versions
- cicero, 5 Nov 2026, Post-train of Qwen3.5-35B-A3B, a mixture-of-experts model with 35B parameters and 3B active. The vision tower is the base model's. Apache 2.0 weights as butler-compute/Cicero, with a quantized copy as butler-compute/Cicero-NVFP4. Release notes
Use cases
What people use Cicero for, one page per pattern.
Support inbox triage with System One models
Send a support ticket to Jev once with every question attached. Category comes back as a selected label, severity and frustration as numbers on scales you wrote, refund intent as a probability. Your code reads those values and decides what happens to the ticket.
Intent and model routing with System One models
One Jev call reads an incoming request and returns its intent as a label plus a difficulty rating on a scale you wrote. Your router reads both numbers and picks the handler: deterministic code, a cheap model, an expensive one, or a human queue.
Agent routing and skill selection with System One models
An agent choosing from a long skill roster reads one truncated line per entry and often loads the wrong thing. Jev ranks every entry in one request and separately answers whether any skill applies at all, so the agent gets a short hint instead of a guess.
Benchmarks
Published scores, each linked to its source. A row marked vendor-run was measured by the vendor itself, so read it as one side's result. The compare hub sets these beside the other models.
| Benchmark | Metric | Score | Run by | Date |
|---|---|---|---|---|
| Jev Decision Index 0.4.1 | Public score | 71.33 score | vendor-run | 5 Nov 2026 |
- Jev Decision Index, Public score: Butler's own run of the public suite with the official kit, from its model card. Cicero has no row on the Jev Decision Index board as of 2026-11-08, so there is no independent score and no private Full score. Area scores per the card: Tools & Automation 87.0, Retrieval & Classification 76.1, Language Understanding 74.6, Arts & Human Taste 64.4, Knowledge & Reasoning 56.1.