Best Decision Models, Ranked
A decision model answers typed questions (choose one option, score on a rubric, yes or no) with a probability for each allowed answer instead of generating text. Models are ranked by their Full score on the Jev Decision Index, a third-party board of 37 benchmarks with private held-out sets, until scores from our own Metatext Decision Benchmark are published.
Ranked list (top 25)
| # | Model | Decision Index | ECE | p50 latency | Per 1k decisions | Params |
|---|---|---|---|---|---|---|
| 1 | vLLM-SR d3 (27B) vLLM Semantic RouterOpen weights | 66.3 | 0.043 | 97 ms | — | 26B |
| 2 | Decider 31B (NVFP4) MapikaOpen weights | 64.6 | 0.024 | 72 ms | — | 33B |
| 3 | Perplexity Decider v1.1 (27B) PerplexityOpen weights | 62.8 | 0.071 | 104 ms | $0.0040 | 28B |
| 4 | Darwin-27B-ZTC FINAL_BenchOpen weights | 62.5 | 0.045 | 104 ms | — | 26B |
| 5 | AnyJev L0 · Gemma-4-31B morriszjmOpen weights | 60.2 | 0.240 | 98 ms | — | 31B |
| 6 | Fastino GLiDE Fastino | 60.2 | 0.040 | 98 ms | $0.030 | 28B |
| 7 | Sieve 27B Sthānika AIOpen weights | 60.2 | 0.065 | 101 ms | — | 28B |
| 8 | Jev TypeSafe AI | 60.1 | 0.074 | 524 ms | $0.0084 | — |
| 9 | Torchcast Decision 27B Torchcast AIOpen weights | 59.9 | 0.024 | 100 ms | — | 28B |
| 10 | Shisa DE-2 (repeat) Shisa.AIOpen weights | 59.8 | 0.207 | 57 ms | — | 26B |
| 11 | Matilda Jev v1.5 MaincodeOpen weights | 59.2 | 0.077 | 89 ms | — | 26B |
| 12 | deck31b (Gemma 4 31B) krishna-gogineni-765Open weights | 59.0 | 0.204 | 219 ms | — | 33B |
| 13 | Kev 27B Jared PalmerOpen weights | 58.8 | 0.022 | 135 ms | — | 28B |
| 14 | Quyet-1.0-Large chinhncOpen weights | 58.7 | 0.078 | 105 ms | — | 33B |
| 15 | RSI-Jev v6.1 VL 27B (effort auto) shgaoOpen weights | 58.7 | 0.077 | 76 ms | — | 28B |
| 16 | decisio 0.6.0 (Gemma 4 31B) aminryOpen weights | 58.5 | 0.142 | 54 ms | — | 31B |
| 17 | Qwen Decision 27B noncausal LoRA jsaurabhOpen weights | 58.5 | 0.051 | 127 ms | — | 28B |
| 18 | garas (Gemma 4 31B) plantwatererOpen weights | 58.3 | 0.049 | 117 ms | — | 31B |
| 19 | Wald-26B-A4B ORG2 AIOpen weights | 58.3 | 0.043 | 90 ms | — | 26B |
| 20 | linne-fern 1.0 (27B) MISA-AICOREOpen weights | 57.8 | 0.079 | 108 ms | — | 28B |
| 21 | Blink v0.3 26B-A4B (NVFP4) Pixilab.aiOpen weights | 57.8 | 0.040 | 25 ms | — | 26B |
| 22 | Decider chat · Gemma-4-31B MapikaOpen weights | 57.6 | 0.047 | 109 ms | — | 33B |
| 23 | SynACK Decide 26B-A4B v7 SkyPantherOpen weights | 57.5 | 0.069 | 129 ms | — | 26B |
| 24 | Surogate Rune 26B-A4B v3 SurogateOpen weights | 57.4 | 0.120 | 121 ms | — | 26B |
| 25 | Xor 26B-A4B v1.1 Juspay TechnologiesOpen weights | 57.1 | 0.058 | 110 ms | — | 26B |
Scores, calibration (ECE) and latency from the Jev Decision Index; cost per 1,000 decisions at 200 input tokens each. Updated daily.
More decision-model rankings
Frequently asked questions
What is the best decision model?
On the Jev Decision Index, vLLM-SR d3 (27B) ranks first with a Full score of 66.3.
How is a decision model different from an LLM?
An LLM generates text. A decision model returns a probability distribution over the allowed answers for each typed question, so its output can be thresholded, audited and calibrated without parsing text.
Is the most accurate decision model the best choice?
Not always. Calibration (how far stated confidence is from the real hit rate), latency and cost per 1,000 decisions often matter as much; see the calibration and cost rankings.