← all repositories
AnotiaWang/awesome-decision-models

A field guide to models that answer with probabilities, not prose

A community-maintained awesome list mapping the young 'decision model' ecosystem — models that return a probability for each answer you declare up front instead of generating text.

★723 stars Python LearningLanguage ModelsAgents
awesome-decision-models
Too new to rank. Velocity needs a few days of history.
not enough history yet
star history

What it does A curated awesome list for ‘decision models’: models that read a state (text, JSON, sometimes images) plus questions whose answers you declare up front, and return a probability for every answer instead of generated text. Questions come in three shapes — Noul (is this statement true), Choice (one of your options), and Score (a level on an ordered rubric). The list spans hosted APIs, open weights, runtimes, SDKs, applications, benchmarks, and papers, in English and Simplified Chinese, with a website.

The interesting bit Per the README, TypeSafe AI introduced the category in September 2026 with Jev and its /v1/systemone API — and already a crowd of hosted APIs and open-weight models, from Cloudflare down to community builders, speak the same request shape. The list is less a directory than a live record of standardization by imitation. It is also honest curation: a standing warning that scores are self-reported on each project’s chosen benchmark and not comparable across entries.

Key highlights

  • Full-stack coverage: hosted APIs (Jev, OpenAI’s Decisions API, Upstage, Liquid AI, Inception), open weights (Cloudflare’s Clef, Perplexity’s pplx-decider, AWS Strands Decider), runtimes, SDKs, apps, benchmarks, and papers.
  • Open weights span 27B fine-tunes down to a 619 MB GGUF that answers Noul questions CPU-only in about 54 ms.
  • Together AI’s entry comes with a walkthrough on training your own decision model for $17.
  • The community corner includes Doom-playing models where each move is one forward pass, and a model trained by a recursively self-improving research system that publishes its failures.
  • Explicitly community-maintained and unaffiliated with any model provider.

Caveats

  • The deliverable is a README and a website — the Python language tag is incidental; there is nothing to install.
  • Several linked entries are beta, experimental, or limited preview, and the category is months old, so expect churn.
  • Benchmark numbers are author-reported on non-shared benchmarks; treat cross-entry comparisons as directional at best.

Verdict Bookmark it if you’re weighing typed decisions for agent guardrails, routing, or classification and want one map of a scattered, fast-moving field. Skip it if you wanted a tool — this is the map, not the territory.

Frequently asked

What is AnotiaWang/awesome-decision-models?
A community-maintained awesome list mapping the young 'decision model' ecosystem — models that return a probability for each answer you declare up front instead of generating text.
Is awesome-decision-models open source?
Yes — AnotiaWang/awesome-decision-models is open source, released under the CC0-1.0 license.
What language is awesome-decision-models written in?
AnotiaWang/awesome-decision-models is primarily written in Python.
How popular is awesome-decision-models?
AnotiaWang/awesome-decision-models has 723 stars on GitHub.
Where can I find awesome-decision-models?
AnotiaWang/awesome-decision-models is on GitHub at https://github.com/AnotiaWang/awesome-decision-models.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.