Tasks and inputs
Cicero is listed for Image-text-to-text, Text classification, Structured Decisions. It accepts text, image and produces decision. Use the publisher’s input format and preprocessing for your chosen deployment.
Cicero is Butler Compute's open-weight MoE decision model: up to 64 typed questions per state, a calibrated probability for every option, no text.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Unverified: registry lists $0.01/1M input, $0.005/1M cache read, free output on Butler's API. Confirm on your account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
What if a decision model cost almost nothing to run per token? Cicero, from Butler Compute, makes that case with architecture. It is post-trained from Qwen3.5 35B-A3B, a mixture-of-experts model with 35B total parameters and only 3B active. Small active compute means fast answers and cheap hosting.
Cicero does not write. You give it a state, which can be text, a chat transcript or up to four images, together with up to 64 typed questions. It returns a calibrated probability for every option of every question.
Each question is answered in one forward pass, read off the scores the model assigns to option letters. There is no sampling loop and no parsing. The supported types:
The vision tower is the base model's own, unchanged by Butler's post-training.
| Item | Detail |
|---|---|
| API model ID | cicero (owned by butler) |
| Hugging Face | butler-compute/Cicero |
| Quantised build | butler-compute/Cicero-NVFP4 |
| Parameters | 35B total, 3B active |
| Context | 131,072 tokens |
| Licence | Apache-2.0 |
| API style | OpenAI-compatible |
Butler's own run on the public Jev Decision Index 0.4.1 suite gives Cicero 71.33, with 41 ms median latency. Note the wording: this is a self-run on the public suite, not a placement on the official board. Treat it as a strong claim to check, not a ranking.
A registry lists $0.01 per 1M input tokens, $0.005 cache reads and free output on Butler's API. Butler's site renders client-side and we could not read a price page, so that figure is unverified; confirm it on your account.
Dates differ as well. Butler's model list shows Cicero created on 1 November 2026, while the Hugging Face repository dates from 5 November. Either way, it is brand new.
Cicero's 255-option limit is its practical edge. Many decision models top out far lower, which forces you to split large taxonomies into hierarchies. If you classify into a long, flat list, such as product categories or intent codes, test that case directly. Feed in a few hundred labelled examples with the full option list, then check top-1 accuracy and whether the probability on wrong answers is honestly low.
Compare it with TypeSafe's Jev on identical payloads. Other open options are gathered under structured decision models.
Compare the model on the work you need to complete.
Cicero is listed for Image-text-to-text, Text classification, Structured Decisions. It accepts text, image and produces decision. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 131.07K context. Test labelled examples and choose thresholds from the cost of each error. Validate probabilities against your application’s outcomes.
The weights are downloadable under Apache-2.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Gemini Decision 1 powers Google's Decisions API, returning calibrated yes/no, choice and rubric answers instead of text, at $0.10 per 1M input tokens.
Perplexity
pplx-decider-flash-v1.3 is Perplexity's small, half-price decision model, distilled from the 27B decider, for high-volume classification and span extraction.
Perplexity
pplx-decider-v1.3 is Perplexity's 27B-class decision model: it returns probabilities for choice, yes/no, score and new span-locate questions, not text.
Morph
Morph Reflex is a family of fast classifier models that label agent turns (jailbreak, looping, frustration and more) with per-class scores.
Sarvam
Sarvam Prism is a 105B decision model that returns calibrated probabilities for yes/no, choice and score questions in English and 10 Indian languages.
levanto
Levanto Sage 2 is listed as a new version of Levanto's hosted Sage decision model; Levanto's own docs do not yet name a version 2.