Juridik · Vals AI
LegalBench
Hur väl en modell klarar korta juridiska resonemangsuppgifter, mest på amerikansk rätt och på engelska.
Placeringar (32, varav 27 i vår lista)
- 1 88.56% ± 0.33[1]
- 2 88.51% ± 0.42[1]
- 3 88.30% ± 0.37[1]
- 4 87.42% ± 0.41[1]
- 5 87.40% ± 0.33[1]
Gemini 3.1 Pro Preview (02/26) Google - 6 87.26% ± 0.42[1]
- 7 87.03% ± 0.37[1] Gemini 3 Pro (11/25) Google
- 8 86.99% ± 0.43[1]
- 9 86.97% ± 0.42[1]
- 9 86.97% ± 0.41[1]
- 11 86.86% ± 0.36[1] Gemini 3 Flash (12/25) Google
- 12 86.70% ± 0.41[1]
- 13 86.52% ± 0.41[1]
GPT 5.5 OpenAI - 14 86.31% ± 0.42[1]
- 15 86.23% ± 0.41[1]GLM LegalZ.ai
- 23 85.26% ± 0.45[1]
- 24 85.25% ± 0.45[1]
- 29 84.84% ± 0.40[1]
- 30 84.74% ± 0.45[1]
- 33 84.39% ± 0.43[1]
- 33 84.39% ± 0.46[1]
- 37 84.23% ± 0.42[1]
- 38 84.22% ± 0.42[1]
- 45 83.93% ± 0.43[1]
- 48 83.76% ± 0.41[1]
- 51 83.61% ± 0.42[1]
- 56 83.28% ± 0.46[1]
- 72 82.43% ± 0.47[1]
- 73 82.36% ± 0.44[1]
- 89 80.32% ± 0.47[1]
- 111 77.71% ± 0.50[1]
- 113 77.13% ± 0.55[1]
Källor
- [1] LegalBench Leaderboard and Methodology ↗Vals AITopplistavals.aiPublicerad 13 okt 2026 Läst 25 okt 2026
- [2] LegalBench ↗Hazy Research, StanfordOfficiell sidahazyresearch.stanford.eduPublicerad okänt datum Läst 25 okt 2026
Historik
- 25 okt 2026Fler placeringar länkade till modeller som lagts till i listan.
- 25 okt 2026Tillagd. Avläst 25 okt 2026.