Jev directory
CommunityJev Models & Reimplementations
Clef Flash on Ollama
Cloudflare's 9B multimodal decision model on Ollama: text, JSON, or images in one non-autoregressive forward pass, Apache-2.0, and the lowest measured latency of any decision model in Cloudflare's Decision Index runs at 38.8 ms median.
- Reading time
- 2 min
- Published by
- Community
- Author
- Cloudflare
- Last verified
- Oct 2026
- Platforms
- CLI
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- 9B decision model from Cloudflare, fine-tuned from Qwen3.5-9B, listed with 22.6K downloads on the Ollama library.
- Scores every option of every question jointly in one non-autoregressive pass, for 1 to 64 named questions per call.
- Reads text, JSON, or base64 PNG, JPEG, and WebP images on 11-12 GB checkpoints with a 256K context window in the listing.
- Decision Index runs put it at 38.8 ms median and 122.4 ms p95 latency; Cloudflare says the 27B Clef is about five times slower.
- Apache 2.0 licence and /v1/systemone support; requires Ollama 0.35.1 or later. Questions: choice, noul, and score.
Watch out
Apache-2.0 weights, but the readme's highlights claim a 64K context while the model listing shows 256K, and the benchmark table mixes Cloudflare's Decision Index latency runs with Typesafe Evals workflow numbers without a shared methodology. Decision models were not in the Ollama CLI or client libraries at capture, so calls go through /v1/systemone or the TypeSafe SDK.
Questions about this entry
Short answers drawn from the linked project or page.
A 9B multimodal decision model from Cloudflare, fine-tuned from Qwen3.5-9B, that turns a state and a schema of typed questions into decisions. It is the smaller, faster sibling of Cloudflare's 27B Clef.
Pull it with ollama pull clef-flash and call the /v1/systemone endpoint, for example curl http://localhost:11434/v1/systemone. Ollama 0.35.1 or later is required, and decision models were not yet exposed through the Ollama CLI at capture.
Apache 2.0, per the model page. The checkpoint is an 11-12 GB download, and the library lists five model entries including clef-flash:latest and clef-flash:9b.
Yes. Base64-encoded PNG, JPEG, or WebP images ride along with the request and are scored jointly with the text state; URLs and data URLs are not supported. The page lists screenshots, receipts, forms, and photos as targets.
Three: choice picks from 2 to 26 described options, noul answers true or false with the probability that the answer is true, and score places an answer on 2 to 26 ordered levels. Every answer comes back with probabilities and a confidence from 0 to 1.
Cloudflare's Decision Index runs put it at 38.8 ms median and 122.4 ms p95 latency, the lowest measured of any decision model there; the 27B Clef is roughly five times slower at 209.3 ms median. Use Clef Flash for latency-critical decisions.
No. The page says confidence runs from 0 to 1 and shows how concentrated the probabilities are, not the chance that the answer is right. If two options tie, the answer follows the model's option order, so put your preferred option first.
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
Bespoke Labs' 9B decision model on Ollama, fine-tuned from Qwen3.5-9B: Apache-2.0, up to 64 typed questions per call, answers in under 100 ms on a MacBook Pro with an M5 Max, and 75.7% mean accuracy across 13 human-labelled datasets.
Jev Models & Reimplementations#community#open-models#local
Community
ActBro's 2B calibrated System 1 decision model on Ollama, built on OpenBMB's MiniCPM5-2B: 78.79% on JevBench among sub-2B open-weight models, 86.41% on the 10-source decision-v7 set, and one output token per decision via /v1/systemone.
Jev Models & Reimplementations#community#open-models#local
Community
A 70.8M-parameter DeBERTa-v3-xsmall cross-encoder that answers typed decision questions in one batched pass with zero generated tokens, shipped with a calibration.json that the engine requires.
Jev Models & Reimplementations#community#open-models#local
Community
A 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 Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
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.