Jamba 1.7 Mini: benchmark results

AI21 Labs' Jamba 1.7 Mini, a hybrid SSM-Transformer model in AI21's Mini tier (12B active of 52B total parameters), with a 256K-token context window for business use cases like function calling. Provider: AI21 Labs. Released 2025-07-03. Access: API.

Aggregate Rating 1458 ± 18, rank #945 of 1656 rated models, from 115 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
RoBench-25 (No Defense Instruction) - Trigger G1Clipped Heidke skill score (0-1; chance-corrected accuracy o100
RoBench-25 (Recovery Instruction) - Trigger C0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o90
RoBench-25 (Recovery Instruction) - Trigger E0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o90
RoBench-25 (Awareness and Recovery Instructions) - Trigger C0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (Awareness and Recovery Instructions) - Trigger E0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (Awareness and Recovery Instructions) - Trigger F0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (No Defense Instruction) - No Trigger1Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (No Defense Instruction) - Trigger A1Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (No Defense Instruction) - Trigger D1Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (No Defense Instruction) - Trigger E1Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (No Defense Instruction) - Trigger F1Clipped Heidke skill score (0-1; chance-corrected accuracy o83.3
RoBench-25 (Recovery Instruction) - No Trigger0.99Clipped Heidke skill score (0-1; chance-corrected accuracy o80

Interactive version: theaggregate.ai/model?slug=jamba-1-7-mini · How It Works · Data refreshed daily, snapshot 2026-10-20.