Tev1 4B Experimental

Together · togethercomputer/tev1-4b-experimental · added 20 Oct 2026

Tev1 4B Experimental by Together ranks #13 of 17 AI decision models overall on DecisionBench: #14 of 17 on Sudoku, #14 of 17 on Rulebook, #11 of 17 on Consistency, #9 of 17 on Odds. View on OpenRouter.

Overall: #13 of 17
Overall score: 40 / 100
Sudoku: #14 of 17
Rulebook: #14 of 17
Consistency: #11 of 17
Odds: #9 of 17

Logical reasoningSudoku#14

Accuracy: 75.4%
Solved: 20/20
Mistakes / puzzle: 16.4

Full Sudoku results →Following written rulesRulebook#14

Accuracy: 81.1%
Confidence gap: 0.103
Cost / 1k tickets: $0.0706

Full Rulebook results →Not getting thrown offConsistency#11

Score: 97.6%
Same if re-asked: 100.0%
Obeyed planted text: 5.0%

Full Consistency results →Weighing riskOdds#9

Score: 38.2%
Value captured: 76.4%
Odds within 2×: 0.0%

Full Odds results →

On each benchmark the best model scores 100 and the worst 0. The overall score averages all 4, so every benchmark counts the same. How ranking works.