Jamba 1.6 Mini: benchmark results
AI21 Labs' Jamba 1.6 Mini, a hybrid SSM-Transformer model with 12B active of 52B total parameters and a 256K-token context window, built for enterprise function-calling and structured output. Provider: AI21 Labs. Released 2025-03-06. Access: API.
Aggregate Rating 1420 ± 25, rank #1147 of 1656 rated models, from 43 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Enkrypt AI - Jailbreak Risk | 10.53 | Risk Score | 43.5 |
| Wolfram LLM Benchmarking Project | 33.2 | Correct Functionality (%) | 33.3 |
| Enkrypt AI - Safety Risk | 32.38 | Risk Score | 31.3 |
| Enkrypt AI - Bias Risk | 87.34 | Risk Score | 24.7 |
| AI for Education Pedagogy - Social studies | 61.82 | Accuracy (%) | 20.3 |
| AI for Education Pedagogy - Technology | 68.87 | Accuracy (%) | 19.3 |
| AA Humanity's Last Exam | 4.31 | Accuracy (%) | 18.4 |
| LegalBench | 69.73 | Score (%) | 17.2 |
| Enkrypt AI - Toxicity Risk | 8.77 | Risk Score | 16.5 |
| AI for Education Pedagogy | 61.07 | Accuracy (%) | 16.4 |
| AI for Education Pedagogy - Primary | 64.32 | Accuracy (%) | 16.4 |
| Vals AI LegalBench | 69.73 | Accuracy (%) | 16 |
Interactive version: theaggregate.ai/model?slug=jamba-1-6-mini · How It Works · Data refreshed daily, snapshot 2026-10-20.