jamba-large-1.7: benchmark results
AI21's hybrid SSM-Transformer MoE (398B total/94B active) with a 256K context, updated for better grounding and instruction-following (July 2025). Provider: AI21 Labs. Released 2025-07-03. Access: API.
Aggregate Rating 1575 ± 12, rank #525 of 1656 rated models, from 316 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - fig_qa | 2.44 | Dataset z-score | 98.8 |
| AGC-Bench - story_generation_rocstories | 2.32 | Dataset z-score | 97.6 |
| UGI Leaderboard | 52.21 | UGI Score | 92.9 |
| AGC-Bench - twistlist | 1.43 | Dataset z-score | 92.7 |
| AGC-Bench - c3_crosstalk | 1.07 | Dataset z-score | 90.2 |
| AGC-Bench - metaphor_generation | 1.05 | Dataset z-score | 90.2 |
| Evals for Every Language - Slovak (sk) | 74.38 | Average Score (%) | 86.4 |
| AGC-Bench - science_analogies | 0.89 | Dataset z-score | 86.2 |
| AA Omniscience - Software Engineering (SWE) - Julia | 28 | Accuracy (%) | 85.2 |
| UGI - Natural Intelligence | 36.65 | NatInt Score | 83.4 |
| AGC-Bench - outline_to_story | 1.02 | Dataset z-score | 82.7 |
| Evals for Every Language - Chinese (zh) | 74.82 | Average Score (%) | 81.8 |
Interactive version: theaggregate.ai/model?slug=jamba-large-1-7 · How It Works · Data refreshed daily, snapshot 2026-10-20.