Jev directory
CommunityJev Models & Reimplementations
Torchcast Decision 27B
A 27B decision model from torchcast-ai that answers choice, noul and score questions together in one forward pass over the TypeSafe-compatible /v1/systemone format; the card reports Decision Index 0.2.1 of 65.00, +7.09 over Jev 1.13, at 33 ms per yes/no question.
- Reading time
- 3 min
- Published by
- Community
- Author
- Torchcast AI
- Last verified
- Oct 2026
Highlights
- Answers choice, noul and score questions together in one forward pass, returning a probability per option via /v1/systemone.
- Card reports Decision Index 0.2.1 of 65.00, +7.09 over Jev 1.13, scored with the official kit across all 155,390 suite rows.
- Latency: 33.0 ms for one yes/no question and 49.0 ms for a three-question request, end to end over HTTP on one H100 80GB in bf16.
- Context up to 262,144 tokens (256K); requests and responses use the TypeSafe-compatible format, so Jev choice/noul/score clients should work unchanged.
- Ships a fast inference package with flash-linear-attention and causal-conv1d kernels and CUDA graphs; without them the server refuses to start on a GPU.
Quickstart
bash
hf download torchcast-ai/torchcast-decision-27b --local-dir torchcast-decision-27b
cd torchcast-decision-27b
pip install -r requirements.txt
python -m torchcast_decision.check .
python -m torchcast_decision.server --model . --port 8090Watch out
Card reviewed 2026-10-17; the 65.00 Decision Index is author-run with the official kit and is not on the public board, and latency figures are not comparable across models because published numbers use different hardware and paths. The base model's licence is cc-by-nc-4.0, the bf16 weights need 54.7 GB and a 64 GB NVIDIA GPU, and scoring split the suite between two GPUs each holding a full model copy to halve wall-clock time.
Questions about this entry
Short answers drawn from the linked project or page.
A 27B decision model from torchcast-ai that answers typed questions about a state: choose one of several options, yes or no, or a rating on an ordinal scale. Every question comes back with a probability for each option, and typical requests are answered in a single forward pass.
Requests and responses use the TypeSafe-compatible /v1/systemone format, so the card says clients written for the choice, noul and score questions of Jev should work without changes. Context runs up to 262,144 tokens, though the published evaluations are text-only.
Decision Index 0.2.1 of 65.00, +7.09 over Jev 1.13, scored with the official kit on all 155,390 suite rows at coverage 1.0; 33.0 ms for one yes/no question and 49.0 ms for a three-question request, mean of 100 after 20 warm-ups on one H100 80GB.
The card says it is ahead on 31 of the 38 index benchmarks, with the largest gains in tool use, POP909-CL, aspect-sentiment extraction, stance, grade-school math, fact verification and contract reasoning. Jev remains ahead on GPQA Diamond, MMLU-Pro and BBH, and its Knowledge area is 5.8 points higher.
Linux with an NVIDIA GPU with at least 64 GB of memory; the bf16 weights are 54.7 GB. The card's fast path needs flash-linear-attention and causal-conv1d, and without those kernels transformers falls back to a much slower path and the server refuses to start on a GPU.
The base model is cc-by-nc-4.0, so non-commercial use. The included torchcast_decision/ inference package is Apache-2.0 and derived from StartLux-Decision's inference code, per its NOTICE; the weights and the package therefore carry different terms.
The card's table lists StartLux-Decision-27B at 63.88, Clef at 61.21 (self-reported by Cloudflare), Jev 1.13 at 57.91, Surogate Rune 26B-A4B v3 at 57.44, Decider chat Gemma-4-31B at 57.33, pplx-decider-v1-27b at 56.40 and Winnow-12B at 50.02.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: ASTGL replay chose local Laya routing over hosted Jev on one Mac Studio setup.
Jmeg8r15 pointsDoes an open-weight decision model beat a hosted one? Jev vs. LayaRead the thread on Hacker NewsReddit: A walkthrough of routing between models with Jev.
X: Standard One SH (StandardThinking)
Facebook: Intake Router keeps diagnosis out of the answer space at the front desk.
FacebookClinicnotes
More Jev resources like this
The Ollama build of iapp's OpenThai-SystemOne: a 0.8B Thai and English decision model whose v0.3 weights are fine-tuned for Ollama's /v1/systemone letter readout, reporting a 74.3 macro across 13 public subsets for Q8_0.
Jev Models & Reimplementations#community#open-models#local
Community
A 27B decision model from StartLux Labs that takes text, JSON or images up to 262,144 tokens and returns a probability per option through the TypeSafe /v1/systemone format, reporting Decision Index 0.2.1 63.88 and 208 of 231 on JevBench public.
Jev Models & Reimplementations#community#open-models#local
Community
The Q8_0 8-bit GGUF build of StartLux-Decision-v1-27B, 28.60 GB for llama.cpp, served through a decision wrapper that the card measures at 100% same decisions as the original and 209 of 231 on JevBench public.
Jev Models & Reimplementations#community#open-models#quantized
Community
The Q8_0 8-bit GGUF build of StartLux-Decision-v1-9B, 9.53 GB for llama.cpp, with the card reporting 100.0% same decisions as the original weights and 200 of 231 correct (86.6%) on JevBench public.
Jev Models & Reimplementations#community#open-models#quantized
Community
A 2B decision model from Strands Agents for agentic workflows: a LoRA adapter on Qwen/Qwen3.5-2B-Base with a readout head that answers noul, choice and score questions with calibrated confidence, reporting 0.7619 on JevBench at 4096 context.
Jev Models & Reimplementations#community#open-models#local
Community
FRIDA-Decisions answers Russian text with structured decisions in a single encoder pass — choice, ordinal score, yes/no or ranking — and scores 0.893 on the 735-item razvilka benchmark, effectively level with TypeSafe Jev's 0.897 (McNemar p = 0.84).
Jev Models & Reimplementations#community#open-models#russian
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.