Jev directory
CommunityJev Models & Reimplementations
PPLX Decider v1 27B
Perplexity AI's 27B decision model fine-tuned from Qwen/Qwen3.8-27B, producing option probabilities through a custom head instead of generated text; the card reports 85.71% overall across 11 benchmarks against Jev's 84.51%.
- Reading time
- 2 min
- Published by
- Community
- Author
- Perplexity AI
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Platforms
- CLI
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Fine-tuned from Qwen/Qwen3.8-27B; answers come from a custom decision head rather than normal token generation.
- Card's 11-benchmark table: 85.71% overall, ahead of Jev's 84.51%, though Jev leads JevBench public hard 73.27% to 70.30%.
- Best in row on FinancialPhraseBank 84.18%, RAGTruth 88.80%, TabFact 90.60% and Circa 89.20%; ties Qwen3.8-27B on ContractNLI at 80.78%.
- Supports text and images: pass images=["screenshot.png"] to predict, or run inference.py with --image screenshot.png.
- Needs Python 3.12+ and a CUDA GPU with room for about 49 GiB of weights plus working memory; the card lists a newer v1.1 checkpoint.
Quickstart
bash
uvx --from huggingface-hub hf download perplexity-ai/pplx-decider-v1-27b inference.py --local-dir .
uv run inference.pyWatch out
Card reviewed 2026-10-17; all 11 benchmark accuracies are self-reported on the card's own table, and the pplx-decider-v1-27b column was measured through the Perplexity API rather than reproduced from these weights. The card itself marks a newer checkpoint, perplexity-ai/pplx-decider-v1.1-27b, as new_version, and the compared Jev column is also a card-supplied number.
Questions about this entry
Short answers drawn from the linked project or page.
Perplexity AI's 27B decision model fine-tuned from Qwen/Qwen3.8-27B. It produces probabilities for typed decisions rather than generated text, supports text and image inputs, and the card measures it across 11 benchmarks through the Perplexity API.
On the card's own 11-benchmark table it scores 85.71% overall against Jev's 84.51%, but Jev leads on WinoGrande (90.70% to 83.30%), BBH (94.27% to 82.80%), JevBench public hard (73.27% to 70.30%), Belebele and TruthfulQA binary.
WinoGrande, FinancialPhraseBank, RAGTruth, JudgeBench, BBH, JevBench public hard, TabFact, ContractNLI, Circa, Belebele and TruthfulQA binary, with overall averages for Jev (84.51%), Qwen3.8-27B (74.76%) and pplx-decider-v1-27b (85.71%).
Python 3.12+ and a CUDA GPU with about 49 GiB of weights plus working memory. Download inference.py with uvx --from huggingface-hub hf download, then uv run inference.py, or use Decider.from_pretrained and model.predict with a typed question.
Yes. Pass images=['screenshot.png'] to model.predict alongside the text, or run uv run inference.py --image screenshot.png. The card includes a choice example about routing a support request and a noul example asking whether a message expresses urgency.
The card says the data builder uses Bespoke Labs' Nimble converters to turn human-labelled public datasets into Choice and Noul decisions, and pins Nimble's published benchmark subsets by record ID and checksum to keep those records out of training.
Apache-2.0 per the card's metadata. The weights are distributed on Hugging Face for local use, and the card asks users of the data pipeline portion to cite Bespoke Labs' Nimble alongside this model.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
X: Standard One SH (StandardThinking)
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A Q8_0 GGUF of knowlet's Gemma-4-12B-Unified-System-One with a native System One decision runtime and an independently fitted GGUF temperature, shipped with an F16 mmproj for the model's native image and audio projections.
Jev Models & Reimplementations#community#open-models#quantized
Community
A local server that reads next-token probabilities from a fine-tuned Spark model and returns Choice, Score, and Noul answers with zero generated tokens.
Jev Models & Reimplementations#community#python#open-models
Community
Rizzo-AI-Academy/rizzo-flowPython
Rizzo-AI-Academy/rizzo-flowPythonAn Apache-2.0 decision engine with a TypeSafe-compatible endpoint, document answers with source passages, and a ViZDoom demo that picks actions from structured game state.
Jev Models & Reimplementations#community#python#local
Community
deepanwadhwa/OpenDecisionPython
deepanwadhwa/OpenDecisionPythonA small distilled sibling of Perplexity's pplx-decider-v1.2: a Qwen3.5-4B decider that reads a state plus typed choice, noul and score questions in one pass, keeping most of the 27B model's Decision Index.
Jev Models & Reimplementations#community#open-models#multimodal
Community
Perplexity's 27B decision model rebuilt on Qwen3.8-27B with a joint request format that answers a whole request's questions in one sequence, reporting 68.09 on Decision Index 0.3 and 77.43 on vision.
Jev Models & Reimplementations#community#open-models#multimodal
Community
A community Q8_0 GGUF of Mapika/decider-0.8b, the smallest System One typed-decision model: one forward pass returns calibrated Choice, Score and noul answers, reporting 0.405 on JevBench hard-public at 920 MB resident.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.