Jev alternatives: 11 decision models compared
Last verified: . How we rank
Hosted TypeSafe Jev is the default System One API for many agent stacks, but it is not the only way to score Choice, Score, and Noul questions in one forward pass. This hub lists eleven comparison pages we maintain in 2026: Laya on the existing learn hub plus ten dedicated /alternatives guides.
The table order is editorial as of 2026-10-12, informed by GitHub stars, Hugging Face downloads, and on-site interest, but not a single-metric sort. Row eleven (Von) is an explicit editorial speed pick.
We label vendor-reported benchmarks separately from independent Decision Index Full scores from the public snapshot regenerated on 2026-10-10.
Quick picks by need
- Open weights you can fine-tune on your GPUs: Convai Laya or Jared Palmer Kev Compare
- Hosted API on infrastructure you already pay for: Cloudflare clef on Workers AI Compare
- Highest public Decision Index score among listed open weights: Perplexity pplx-decider v1.1 (27B) Compare
- Editorial speed pick for local routing: wfzyx Von (row 11, adoption plus curator pick) Compare
- Stay on TypeSafe billing and gateway integrations: Keep hosted Jev Compare
Eleven comparison pages (editorial order, informed by adoption)
| # | Model | Maker | Open weights | Hosted API | Jev API compatible | Sizes | Inputs | Decision Index | Adoption (dated) | Guide |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Laya | Convai Innovations | Apache 2.0 checkpoints | Self-host only | Conceptual (Router API, not identical wire) | English ~421M, multilingual ~322M (HF model card) | Text / JSON state | Laya vs Jev benchmarks guide (vendor vs independent splits) | 32,393 GitHub stars (2026-10-12); 44,088 HF downloads, apache-2.0 (2026-10-12) | Open |
| 2 | Cloudflare clef | Cloudflare | Weights on HF; serve via Workers AI | $0.24 per M input tokens (vendor list, 2026-10-12) | Workers AI System One style HTTP | clef and clef-flash SKUs | Text state | 53.08 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 14,351 HF downloads, apache-2.0 (2026-10-12) | Open |
| 3 | Kev | Jared Palmer | Apache 2.0 | Self-host (kev.serve) | POST /v1/systemone compatible | 0.8B to 27B | Text state | 58.78 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 8,952 GitHub stars (2026-10-12); 21,656 HF downloads (2026-10-12) | Open |
| 4 | OpenAI Decisions API | OpenAI | Hosted beta only | gpt-6-luna (vendor, beta): $0.10 per M input tokens (vendor list, 2026-10-12) | Decisions API (beta contract) | gpt-6-luna class (vendor) | Structured decisions payload | Not on public Decision Index board in our snapshot | Vendor beta (no public GitHub stars) | Open |
| 5 | Perplexity pplx-decider | Perplexity | Apache 2.0 weights | $0.02 per M input tokens (vendor list, 2026-10-12) | System One style serving options | 27B multimodal decider | Text and image (vendor) | 62.75 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 2,221 HF downloads, apache-2.0 (2026-10-12) | Open |
| 6 | Mapika decider | Mapika | Apache 2.0 | Self-host (decider.serve) | POST /v1/systemone compatible | 0.8B to 35B-A3b | Text state | 25.8 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 1,127 GitHub stars (2026-10-12); 331,532 HF downloads (2026-10-12) | Open |
| 7 | GLiNER2.5 Decide | Fastino | Apache 2.0 | Self-host | Decision head on GLiNER family | Decide variant checkpoints | Text spans / criteria | 9.24 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 90,464 HF downloads, apache-2.0 (2026-10-12) | Open |
| 8 | SemIf | Theo Lee CJ | Open research weights | Self-host (openjev.com demos) | System One shaped server | Multiple community bridges | Text / structured state | Not listed for SemIf in Decision Index snapshot | 4,785 GitHub stars (2026-10-12) | Open |
| 9 | CLM 8B | Contrastive LM | Apache 2.0 | Self-host | Qwen3 heads for decisions | 8B class | Text state | 6.24 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 2,981 GitHub stars (2026-10-12); 4,367 HF downloads, apache-2.0 (2026-10-12) | Open |
| 10 | Liquid d1 | Liquid AI | LFM license weights | Self-host | Decision tuned LFM stack | d1-3B featured | Text state | 40.08 (Decision Index 0.3.1, snapshot 2026-10-10; independent harness, not vendor-only) | 10,843 HF downloads, lfm-1.0 (2026-10-12) | Open |
| 11 | Von | wfzyx | Apache 2.0 | Self-host | System One shaped local server | 395M class encoder (vendor README) | Text state | Not listed on Decision Index public board in our snapshot | 883 GitHub stars (2026-10-12); 106,915 HF downloads, apache-2.0 (2026-10-12) | Open |
How we ranked
The table order is editorial as of 2026-10-12. We use GitHub stars, Hugging Face downloads, and on-site interest as signals, but we do not sort purely by any one metric (downloads alone would reorder several rows). Row eleven (Von) is an explicit editorial speed pick. This is not a quality score. Re-check monthly and keep retired pages linked from the hub when a model drops out of the table.
Hosted APIs (OpenAI Decisions API, Perplexity hosted decider, Cloudflare Workers AI) sit beside open weights in the table but are grouped separately in prose when you need a closed vendor endpoint.
Hosted vs self-hosted
Pick hosted Jev or a vendor API when you want keys, billing, and gateway integrations without running GPUs. Pick open weights when data must stay in your VPC, when you fine-tune checkpoints, or when unit economics favor amortized hardware. See where to run Jev and the System One primer.
Switching from Jev
Compatible servers keep the Choice, Score, and Noul request shape: shared state, typed questions, explicit options. What changes is calibration, option limits, context windows, and thresholds. Many teams shadow-run a candidate model before cutover. See migration and coexistence for patterns.
More open alternatives
Runners-up with strong download counts include Von, NanoJev, and AnyJev. Browse open models on Explore.
For small or self-hosted checkpoints, see Liquid d1 vs Jev and Kev vs Jev.
FAQ
- What counts as a Jev alternative?
Any engine that answers typed decision questions (Choice, Score, Noul) over structured state without generating open-ended chat. That includes hosted System One APIs, open checkpoints with compatible HTTP servers, and research stacks like SemIf. General LLM classifiers are out of scope for this hub.
- Why is Laya first in the table if this site covers Jev?
Row one is an editorial choice: Laya has the largest GitHub and Hugging Face footprint among open decision models we track, and many teams evaluate it alongside hosted Jev. Our Laya vs Jev compare guide is independent writing; we are not Convai or TypeSafe.
- How should I read Decision Index numbers?
Decision Index is a third-party harness (edition 0.3.1) that scores public benchmark slices. It is useful for directional ranking but not a substitute for your own labelled data. Vendor blog tables may use different splits; we mark those as vendor-reported in each alternative page.
- Why is Von in the table as a runner-up?
Von sits at row eleven as an editorial speed pick with strong Hugging Face downloads, not because it outranked Liquid d1 on every signal we watch. Read the Von vs Jev page before you treat it as a full migration target.
- Can I mix hosted Jev with an open model?
Yes. A common pattern is a local open model for high-volume pre-filters and hosted Jev for escalation, large option sets, or compliance workflows. See migration sections on Kev, decider, and Laya pages for minimal SDK diffs.
- Are Cloudflare clef and clef-flash the same product?
They are sibling Workers AI models with different price points and Decision Index scores. This hub's clef page focuses on the standard clef SKU; clef-flash is noted in facts when you need a cheaper flash tier.
- Is OpenAI Decisions API production ready?
OpenAI describes Decisions API as a public beta with a Luna-class decision model. Treat pricing, model names, and SLAs as vendor beta terms until your account team confirms otherwise.
- How often is this hub updated?
We refresh GitHub stars, Hugging Face downloads, hosted price list entries, and Decision Index snapshot scores on the date shown as last verified. Always confirm contracts and model cards before you ship.