Jared PalmerDecision matchup
Kev 0.8Bvs.Kev 4B
Which model makes better decisions?
Kev 4B does better in independent tests, while Kev 0.8B is more popular.
Pick Kev 0.8B for
Pick Kev 4B for
Shared tests
Independent evaluators and leaderboards only, the newest shared snapshot per evaluator. With 95 % ranges a lead counts when they do not overlap; for hit rates, when it holds in a 95 % test. Without any uncertainty figure at most “slightly ahead”.
Jev Decision Index
Kev 4B aheadKev 4B is clearly ahead in 20 of 37 subtests and slightly in 12, Kev 0.8B in none. 5 are level.
Home appliances · Decision Index 0.4.1
88 tasks, as of October 28, 2026
SATA-Bench · Decision Index 0.4.1
1,650 tasks, as of October 28, 2026
GSM8K · Decision Index 0.4.1
2,638 tasks, as of October 28, 2026
Show all 37 subtests
| Subtest | Kev 0.8B | Kev 4B | Verdict |
|---|---|---|---|
| Home appliances · Decision Index 0.4.1 | 0.0 | 14.8 | Kev 4B ahead |
| SATA-Bench · Decision Index 0.4.1 | 9.0 | 28.8 | Kev 4B ahead |
| GSM8K · Decision Index 0.4.1 | 19.0 | 46.5 | Kev 4B ahead |
| MMLU-Pro · Decision Index 0.4.1 | 22.3 | 50.7 | Kev 4B ahead |
| New Yorker · Decision Index 0.4.1 | 27.8 | 52.3 | Kev 4B ahead |
| HellaSwag · Decision Index 0.4.1 | 42.9 | 78.9 | Kev 4B ahead |
| When2Call · Decision Index 0.4.1 | 43.2 | 66.7 | Kev 4B ahead |
| BBH · Decision Index 0.4.1 | 41.4 | 62.5 | Kev 4B ahead |
| BPoMP · Decision Index 0.4.1 | 47.7 | 65.1 | Kev 4B ahead |
| WinoGrande · Decision Index 0.4.1 | 52.8 | 70.2 | Kev 4B ahead |
| BFCL · Decision Index 0.4.1 | 70.5 | 93.1 | Kev 4B ahead |
| CRUXEval · Decision Index 0.4.1 | 37.7 | 48.1 | Kev 4B ahead |
| Habermas · Decision Index 0.4.1 | 33.4 | 42.0 | Kev 4B ahead |
| CLadder · Decision Index 0.4.1 | 50.5 | 63.1 | Kev 4B ahead |
| API-Bank · Decision Index 0.4.1 | 43.7 | 53.3 | Kev 4B ahead |
| HoVer · Decision Index 0.4.1 | 51.0 | 59.7 | Kev 4B ahead |
| MuSR · Decision Index 0.4.1 | 48.9 | 56.3 | Kev 4B ahead |
| PhishNChips · Decision Index 0.4.1 | 51.7 | 55.3 | Kev 4B ahead |
| Humicroedit · Decision Index 0.4.1 | 50.9 | 54.4 | Kev 4B ahead |
| cfcolor · Decision Index 0.4.1 | 55.4 | 57.9 | Kev 4B ahead |
| CLINC150 · Decision Index 0.4.1 | 15.0 | 75.0 | Kev 4B slightly ahead |
| iSarcasmEval · Decision Index 0.4.1 | 19.9 | 46.8 | Kev 4B slightly ahead |
| Amazon ESCI · Decision Index 0.4.1 | 22.9 | 39.7 | Kev 4B slightly ahead |
| ContractNLI · Decision Index 0.4.1 | 40.9 | 64.4 | Kev 4B slightly ahead |
| RAGTruth · Decision Index 0.4.1 | 46.2 | 67.5 | Kev 4B slightly ahead |
| ANLI · Decision Index 0.4.1 | 33.6 | 48.1 | Kev 4B slightly ahead |
| NLI4CT · Decision Index 0.4.1 | 55.9 | 74.6 | Kev 4B slightly ahead |
| VAST · Decision Index 0.4.1 | 40.6 | 52.4 | Kev 4B slightly ahead |
| BRIGHT · Decision Index 0.4.1 | 31.8 | 39.0 | Kev 4B slightly ahead |
| FinEntity · Decision Index 0.4.1 | 71.4 | 87.1 | Kev 4B slightly ahead |
| ToolRet · Decision Index 0.4.1 | 51.7 | 61.0 | Kev 4B slightly ahead |
| BANKING77 · Decision Index 0.4.1 | 77.6 | 82.9 | Kev 4B slightly ahead |
| HLE · Decision Index 0.4.1 | 13.8 | 10.2 | level |
| POP909 · Decision Index 0.4.1 | 1.5 | 1.2 | level |
| GPQA Diamond · Decision Index 0.4.1 | 33.2 | 37.8 | level |
| ChessBench · Decision Index 0.4.1 | 7.9 | 8.9 | level |
| ACOS · Decision Index 0.4.1 | 9.9 | 9.8 | level |
Source · latest October 28, 2026
Kevala decision benchmark report
Kev 4B aheadKev 4B is clearly ahead in 1 of 3 subtests and slightly in 1, Kev 0.8B in none. 1 is level.
Kevala semif-authored144 decision accuracy
95 % ranges 64.6 to 77.3 and 85.9 to 93.1, 1,296 tasks, as of October 23, 2026
Kevala semif-perturbations108 decision accuracy
95 % ranges 57.4 to 84.0 and 81.2 to 98.8, 972 tasks, as of October 23, 2026
Kevala kevala-authored36 decision accuracy
95 % ranges 79.6 to 97.2 and 82.4 to 98.2, 324 tasks, as of October 23, 2026
Source · latest October 23, 2026
Duan et al., General Decision Models: Benchmarking and Insights Beyond Jev
Kev 4B aheadJEVal aida-testc accuracy
200 tasks, as of October 22, 2026
Source · latest October 22, 2026
JevAny external decision-suite evaluation score table
Kev 4B aheadMMLU-Pro, ekzhang external test panel (1,000 questions)
1,000 tasks, as of October 17, 2026
Source · latest October 17, 2026
Vendor figures
Figures from vendor reports that cover both models. They do not count toward the verdict.
Show 2 vendor figures
| Figure | Source | Kev 0.8B | Kev 4B |
|---|---|---|---|
| 14-public-dataset calibration error, per-dataset average (lower is better) | Jared Palmer, Introducing Kev | 0.10 | 0.10 |
| 14-public-dataset chance-corrected generalization index | Jared Palmer, Introducing Kev | 23.30 | 38.00 |