System One models
21 models measured on both suites and 8 Jev-shaped checkpoints surveyed but not yet run. Open a model for its results on every suite, every configuration and every question, plus the numbers its authors and third parties have published.
Measured
| Model | Mechanism | Size · license | typed-decisions | Metatext |
|---|---|---|---|---|
| Jev jev-1.13.0 | Hosted System One API | — · Proprietary (hosted API) | 0.726 | 0.863 |
| LAYA convaiinnovations/laya | Encoder with a decision head (System One) | 421M · Apache 2.0 | 0.343 | 0.520 |
| LAYA typed-decisions convaiinnovations/laya-typed-decisions | Encoder with a decision head (System One) | 421M · Apache 2.0 | 0.714 | 0.475 |
| LAYA multilingual convaiinnovations/laya-multilingual | Encoder with a decision head (System One) | 322M · Apache 2.0 | 0.345 | 0.431 |
| Von wfzyx/von-1.0 | NLI entailment head, one hypothesis per option | 395M · Apache 2.0 | 0.461 | 0.509 |
| OpenJEV Qwen2.5-0.5B-Instruct (bf16) via open-jev | LLM log-probability scoring (Jev mechanism) | 0.5B · Apache 2.0 | — | 0.427 |
| GLiNER2 large fastino/gliner2-large-v1 | Schema-conditioned encoder (extraction + classification) | 340M · Apache 2.0 | 0.486 | 0.716 |
| GLiNER2 base fastino/gliner2-base-v1 | Schema-conditioned encoder (extraction + classification) | 205M · Apache 2.0 | — | 0.631 |
| GLiNER2 multi fastino/gliner2-multi-v1 | Schema-conditioned encoder (extraction + classification) | 205M · Apache 2.0 | — | 0.464 |
| GLiNER 2.5 base fastino/gliner2.5-base-v1 | Schema-conditioned encoder (extraction + classification) | 194M · Apache 2.0 | 0.398 | 0.643 |
| GLiNER 2.5 small fastino/gliner2.5-small-v1 | Schema-conditioned encoder (extraction + classification) | 74M · Apache 2.0 | — | 0.574 |
| GLiNER 2.5 multi fastino/gliner2.5-multi-v1 | Schema-conditioned encoder (extraction + classification) | 287M · Apache 2.0 | — | 0.494 |
| GLiGuard 300M fastino/gliguard-LLMGuardrails-300M | Schema-conditioned encoder (extraction + classification) | 300M · Apache 2.0 | — | 0.537 |
| GLiNER multitask v1.0 knowledgator/gliner-multitask-v1.0 | Span extractor pointed at labels | ~200M · Apache 2.0 | — | 0.378 |
| GLiClass large v3.0 knowledgator/gliclass-large-v3.0 | Label-relevance classifier | 439M · Apache 2.0 | 0.446 | 0.642 |
| GLiClass base v3.0 knowledgator/gliclass-base-v3.0 | Label-relevance classifier | 187M · Apache 2.0 | — | 0.606 |
| GLiClass modern large v3.0 knowledgator/gliclass-modern-large-v3.0 | Label-relevance classifier | 399M · Apache 2.0 | — | 0.602 |
| GLiClass instruct large v1.0 knowledgator/gliclass-instruct-large-v1.0 | Label-relevance classifier | ~440M · Apache 2.0 | — | 0.545 |
| GLiClass instruct base v1.0 knowledgator/gliclass-instruct-base-v1.0 | Label-relevance classifier | ~190M · Apache 2.0 | — | 0.559 |
| GLiClass multilang ultra knowledgator/gliclass-multilang-ultra | Label-relevance classifier | — · Apache 2.0 | 0.395 | 0.628 |
| GLiClass edge v3.0 knowledgator/gliclass-edge-v3.0 | Label-relevance classifier | 33M · Apache 2.0 | — | 0.418 |
Surveyed, not yet run
A decoder-backbone System One reproduction with a proper-scoring-rule SFT stage and a calibration-aware RL stage; serves the /v1/systemone request shape.
Single-pass encoder with a per-option head over state and all questions in one sequence; 512-token context with the state cut to 256 tokens.
Von's mechanism trained deliberately for the decision task, small and multilingual, with a fitted temperature and the card's own negative results.
A reproduction of the architecture described in "Jev's Architecture Unmasked"; the only one that measured the real Jev on its own frozen dev sets and published a research log.
Scores the literal Jev question JSON as the second sequence, so criteria are read at inference.
Pointwise scorer trained on 46k System One questions across nine families; reports its calibration separately from its accuracy.
Pointwise scorer with a CLI and an "automate if p ≥ 0.9588" gate calibrated to at most 5% error.
A GLiClass-mechanism model with an RLCD-style objective; small enough to run in the browser (ONNX + WebGPU demo).