View as Markdown

Dimmed sections with a dashed outline are placeholders that show the design. Everything at full strength is real.

Latest

Newest first. Scores are the editor’s, and the editor’s score is the headline figure. Upvotes and ratings from readers are counted apart, on each item. How ranking works →

1
REAX 1.1: a 27B decision model fine-tuned from pplx-decider
huggingface.co§ Decisions in the wildREAX (decisionmodels)21 Oct
makerREAX tasktyped decisions
Hype gap ▼ −0.8 under
Shipped
Real3.5
Useful3.0
Notable2.5
Novel2.5
AI first pass
2
Decision4Rec releases Kev models that rerank recommender candidates
huggingface.co§ Decisions in the wildMarco Polignano (m-polignano)19 Oct
makerMarco Polignano taskrecommender candidate ranking
Hype gap ▬ 0.0 level
Shipped
Real3.5
Useful3.0
Notable2.5
Novel4.0
AI first pass
3
AccuKnox Inc. releases Guard 1.1, an Apache-2.0 guardrail for agent inputs and prompt injection
huggingface.co§ Decisions in the wildAccuKnox Inc. (Accuknoxtechnologies)16 Oct
makerAccuKnox Inc.taskprompt injection defence taskcontent moderation
Hype gap ▼ −1.0 under
Shipped
Real4.0
Useful3.5
Notable2.5
Novel3.0
AI first pass
4
Maincode releases Matilda Decision v1, a 27B multimodal decision model
huggingface.co§ Decisions in the wildMaincode (Maincode)15 Oct
makerMaincode tasktyped decisions taskmultimodal classification
Hype gap ▼ −0.5 under
Shipped
Real3.5
Useful3.5
Notable3.0
Novel3.0
AI first pass
5
decisionmodels releases s1-llm-auto-router, a 183 MB CPU router under MIT
huggingface.co§ Decisions in the wilddecisionmodels (decisionmodels)15 Oct
makerdecisionmodels taskLLM request routing
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel3.0
AI first pass
6
Liquid AI releases LFM2.5-24B-A2B, a 24B on-device reasoning MoE with 2B active
huggingface.co§ Open weightsLiquid AI (LiquidAI)12 Oct
makerLiquid AI taskon-device agents tasktool use
Hype gap ▼ −0.2 under
Shipped
Real4.0
Useful3.5
Notable3.5
Novel3.5
AI first pass
7
StartLux releases Decision v1.1, from 0.8B to 27B, with non-commercial weights
huggingface.co§ Decisions in the wildStartLux (startlux-models)11 Oct
makerStartLux tasktyped decisions
Hype gap ▼ −0.5 under
Shipped
Real3.5
Useful3.5
Notable3.0
Novel3.0
AI first pass
8
AI Singapore releases Gemma-SEA-LION-Guard v2, MIT-licensed safety models for Southeast Asian languages
huggingface.co§ Decisions in the wildAI Singapore (aisingapore)11 Oct
makerAI Singapore taskcontent moderation taskaction governance
Hype gap ▼ −0.2 under
Shipped
Real4.0
Useful3.5
Notable3.5
Novel3.5
AI first pass
9
Microsoft releases Decision-1, a 9B model that scores fixed answer options
commandline.microsoft.com§ Decisions in the wildMicrosoft9 Oct
makerMicrosoft taskdecision scoring taskrouting and classification
Hype gap ▼ −0.2 under
Shipped
Real3.5
Useful4.0
Notable4.0
Novel3.0
AI first pass
10
Amazon publishes Strands Decider 2B, a small decision model for agent workflows
huggingface.co§ Decisions in the wildAmazon (amazon)9 Oct
makerAmazon taskagent routing taskguardrails
Hype gap ▼ −0.5 under
Shipped
Real4.0
Useful3.5
Notable3.5
Novel3.0
AI first pass
11
Reflection releases Beam, a 501B open-weights MoE under Apache 2.0
huggingface.co§ Open weightsReflection (reflection)7 Oct
makerReflection tasksoftware engineering taskagentic tool use
Hype gap ▼ −0.2 under
Shipped
Real4.0
Useful3.5
Notable4.0
Novel3.0
AI first pass
12
OpenAI opens its Decisions API in public beta, built on GPT-6 Luna
developers.openai.com§ Decisions in the wildOpenAI6 Oct
makerOpenAI tasktyped decisions taskrouting and classification
Hype gap ▼ −0.5 under
Shipped
Real4.0
Useful4.0
Notable4.0
Novel3.0
AI first pass
13
Torchcast AI releases Torchcast Decision 27B, built on StartLux weights
huggingface.co§ Decisions in the wildTorchcast AI (torchcast-ai)3 Oct
makerTorchcast AI tasktyped decisions
Hype gap ▼ −0.8 under
Shipped
Real3.5
Useful3.0
Notable2.5
Novel2.5
AI first pass
14
Aleph Alpha releases Kolibri 1, a 78B German and English reasoning MoE under Apache 2.0
huggingface.co§ Open weightsAleph Alpha GmbH (Aleph-Alpha)3 Oct
makerAleph Alpha GmbH taskGerman-language assistants taskretrieval-augmented generation
Hype gap ▼ −0.5 under
Shipped
Real4.0
Useful3.5
Notable3.5
Novel3.0
AI first pass
15
SberAI releases FRIDA-Decisions, a Russian and English encoder for structured decisions
huggingface.co§ Decisions in the wildSberAI (ai-forever)2 Oct
makerSberAI tasktyped decisions taskRussian-language classification
Hype gap ▼ −0.5 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel3.5
AI first pass
16
H2O.ai releases H2O Lightning 4B, an Apache-2.0 decision model with GGUF builds
huggingface.co§ Decisions in the wildH2O.ai (h2oai)2 Oct
makerH2O.ai tasktyped decisions
Hype gap ▼ −1.0 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel2.5
AI first pass
17
Perplexity releases pplx-decider, an Apache-2.0 27B decision model
huggingface.co§ Decisions in the wildPerplexity (perplexity-ai)1 Oct
makerPerplexity tasktyped decisions taskclassification
Hype gap ▼ −0.2 under
Shipped
Real3.5
Useful3.5
Notable3.5
Novel3.0
AI first pass
18
Fish Audio releases Drama 3, a 35B mixture-of-experts TTS for multi-character scenes
huggingface.co§ Open weightsFish Audio (fishaudio)1 Oct
makerFish Audio taskmulti-speaker speech taskvoice cloning
Hype gap ▬ 0.0 level
Shipped
Real3.5
Useful3.5
Notable3.5
Novel3.5
AI first pass
19
vLLM Semantic Router team releases Decision 2.0 open decision models
huggingface.co§ Decisions in the wildvLLM Semantic Router (vllm-sr)29 Sep
makervLLM Semantic Router taskrequest routing tasktyped decisions
Hype gap ▼ −0.5 under
Shipped
Real4.0
Useful3.5
Notable3.5
Novel3.0
AI first pass
20
An independent developer releases Jev-Style 2B Decision v3, a 2B model with a free hosted API
huggingface.co§ Decisions in the wildchaoliang yan (chaoliangUNSW)27 Sep
makerchaoliang yan tasktyped decisions
Hype gap ▼ −0.8 under
Shipped
Real3.5
Useful2.5
Notable2.0
Novel2.5
AI first pass
21
Shanghai AI Laboratory releases Intern-Decision, multimodal decision models from 0.8B to 4B
huggingface.co§ Decisions in the wildShanghai AI Laboratory (internlm)26 Sep
makerShanghai AI Laboratory tasktyped decisions taskmultimodal classification
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel3.0
AI first pass
22
Xiaomi MiMo releases MiMo-V2.6-Pro-RL, a 1.02T omnimodal MoE under MIT
huggingface.co§ Open weightsXiaomi MiMo (XiaomiMiMo)21 Sep
makerXiaomi MiMo taskcoding agents taskreinforcement learning
Hype gap ▬ 0.0 level
Shipped
Real3.5
Useful3.5
Notable3.5
Novel3.5
AI first pass
23
distil labs releases an invoice-decision model that approves or holds vendor invoices
huggingface.co§ Decisions in the wilddistil labs (distil-labs)21 Sep
makerdistil labs taskinvoice approval taskthree-way match
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.0
Notable2.5
Novel3.0
AI first pass
24
JetBrains releases Mellum2.1 Thinking, a 12B coding MoE with 2.5B active, under Apache 2.0
huggingface.co§ Open weightsJetBrains (JetBrains)20 Sep
makerJetBrains taskcoding tasksoftware engineering agents
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel3.0
AI first pass
25
Jared Palmer releases Kev-9B, an Apache-2.0 LoRA decision model on Qwen3.5-9B
huggingface.co§ Decisions in the wildJared Palmer (jaredpalmer)20 Sep
makerJared Palmer tasktyped decisions taskdocument classification
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.5
Notable3.0
Novel3.0
AI first pass
26
Convai Innovations releases laya-typed-decisions, a 0.4B ModernBERT decision model
huggingface.co§ Decisions in the wildConvai Innovations (convaiinnovations)18 Sep
makerConvai Innovations tasktyped decisions
Hype gap ▼ −0.8 under
Shipped
Real4.0
Useful3.0
Notable2.5
Novel3.0
AI first pass
27
TypeSafe AI launches Jev, a hosted model for typed software decisions
typesafe.ai§ Decisions in the wildTypeSafe AI (typesafe-ai)15 Sep
makerTypeSafe AI tasktyped decisions taskrouting and classification
Hype gap ▲ +0.8 over
Pending
Real3.0
Useful4.0
Notable4.5
Novel4.0
AI first pass
28
Yandex releases AliceAI-Foundation-80B-A3B-Base, an Apache 2.0 base MoE for Russian and English
huggingface.co§ Open weightsyandex (yandex)12 Sep
makeryandex taskRussian-language modelling taskfine-tuning base
Hype gap ▬ 0.0 level
Shipped
Real4.0
Useful3.0
Notable3.5
Novel3.5
AI first pass
29
DeepSeek releases V4.1-Flash, a multimodal MoE with a much smaller KV cache, under MIT
huggingface.co§ Open weightsDeepSeek-AI (deepseek-ai)10 Sep
makerDeepSeek-AI tasklong-context inference taskagentic workloads
Hype gap ▲ +0.2 over
Shipped
Real4.0
Useful3.5
Notable4.0
Novel4.0
AI first pass
30
OpenBMB releases MiniCPM5-2B, a dense 2B model with open training data, under Apache 2.0
huggingface.co§ Open weightsOpenBMB (openbmb)6 Sep
makerOpenBMB taskon-device inference tasktool use
Hype gap ▼ −0.8 under
Shipped
Real4.5
Useful3.5
Notable4.0
Novel2.5
AI first pass

MoreShipped Open weights Decisions in the wild Roundup Wiki

Clusters

Emerging clusters this month

Clusters found by the topic model, named by AI

Decision models
▲ Growing20 itemsdecision, public, pass, apache20, reports
Accepted by AI
Open-weights language models
▼ Cooling21 itemslicence, reasoning, apache, repo, creation
Accepted by AI
Safety and guard models
9 itemssafety, guard, nemotron, languages, licence
Accepted by AI
Open text-to-speech models
21 itemsvoice, cloning, tts, english, audio
Accepted by AI

8 wks

Change compares the items in the last 4 weeks with the 4 before. No percent when those 4 weeks hold fewer than 5 items or the topic has under 8 weeks of history. Under 4 weeks it is new.

From the wiki · concept of the week

typed decisions

Decision models in Shipped's items score or pick among a fixed set of answer options instead of writing prose. Examples are Jev from TypeSafe AI, pplx-decider from Perplexity, Kev-9B and Matilda Decision v1.

Edited by AI on 2026-10-22

14 items, 14 makers

Rising starts counting once readers vote. Updated every 6 hours.

Rising

Votes / hr, 6-hour windowPlaceholder

Kiln-3B: a small open model for structured extraction, MIT licence▲ +46/hhub.example · Document AI
Port of Calder publishes its container-inspection risk model and false-positive rate▲ +31/hportofcalder.example · Decision models
Quillwork’s contract redlining leaves beta in its Word add-in▲ +22/hquillwork.example · Document AI
Wren TTS weights are out, under a non-commercial licence▲ +19/hcode.example · Speech
Fieldnote ships offline meeting summaries on recent Pixel phones▲ +14/hfieldnote.example · Speech

How to read a row

Shipped: Publicly available. An editor confirmed it, or 3 signed-in readers reported using it.
Verified: Shipped, and an editor checked the evidence.
Pending: Claimed. Nobody has confirmed it yet.
Vapor: Announced, demoed or waitlisted. Not something you can use.

Scores on seeded items are an AI-assisted first pass from the linked sources; human editors re-score them. The headline figure is the editor’s score. Upvotes (readers who want it seen) and ratings (readers who scored it) are counted apart on each item, and ratings join the ranking only once 5 or more readers have rated. How ranking works → How we score →