Jamba 1.5 Large — benchmark results
AI21's 398B-total/94B-active hybrid SSM-Transformer MoE (August 2024) with a 256K effective context, released under the Jamba Open Model License. Provider: AI21 Labs. Released 2024-08-22. Access: API.
Unified ELO 1483 ± 18, rank #866 of 1776 rated models, from 27 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| RULER | 96 | Avg Accuracy (%) | 100 |
| HELM Lite | 68.57 | Mean win rate (self-reported) | 75.3 |
| HELM (Stanford) | 63.7 | Mean Win Rate (%) | 67.8 |
| HELM NaturalQuestions (Closed) | 39.37 | F1 (%) | 62.2 |
| HELM WMT 2014 | 20.29 | BLEU-4 (%) | 61.1 |
| HELM NaturalQuestions (Open) | 71.79 | F1 (%) | 52.2 |
| BenchTable | 39.8 | Total Score (%) | 37.5 |
| ProLLM - Q&A Assistant | 91.3 | Score (%) | 33.3 |
| Chatbot Arena (Text) | 1289 | Elo | 32.6 |
| Vals AI LegalBench | 74.16 | Accuracy (%) | 24 |
| HELM NarrativeQA | 66.36 | F1 (%) | 23.3 |
| AA MATH-500 | 60.6 | Accuracy (%) | 21.3 |
Interactive version: theaggregate.ai/model?slug=jamba-1-5-large · How the rankings work · Data refreshed daily, snapshot 2026-07-22.