Jamba-v0.1 — benchmark results

AI21's Apache-2.0 hybrid Mamba-Transformer MoE (52B total/12B active), the first production-grade SSM-based model, with a 256K context. Provider: AI21 Labs. Released 2024-03-28. Access: API.

Unified ELO 1464 ± 15, rank #936 of 1776 rated models, from 108 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
EuroEval Swedish NLU - Swerec81.11Sentiment classification Score (%)99.8
EuroEval German NLU - Germanquad64.59Reading comprehension Score (%)88.4
EuroEval Icelandic NLU - NQII55.84Reading comprehension Score (%)86.9
EuroEval Finnish NLU - Tydiqa FI71.08Reading comprehension Score (%)85.9
EuroEval Dutch NLU - DBRD91.36Sentiment classification Score (%)85.7
EuroEval Portuguese NLU - SST-2 PT83.67Sentiment classification Score (%)85.5
EuroEval Finnish NLU - Scandisent FI92.03Sentiment classification Score (%)84.8
EuroEval French NLU - FQuAD71.56Reading comprehension Score (%)83.6
EuroEval Portuguese NLU - MultiWikiQA PT74.65Reading comprehension Score (%)83.2
EuroEval Portuguese NLU - ScaLA PT26.68Linguistic acceptability Score (%)81.4
EuroEval Portuguese NLU56.3NLU Average Score (%)79.7
EuroEval Spanish NLU - MLQA ES64.33Reading comprehension Score (%)78.5

Interactive version: theaggregate.ai/model?slug=jamba-v0-1 · How the rankings work · Data refreshed daily, snapshot 2026-07-22.