AI21: Jamba Large 1.7

AI21's Jamba Large 1.7, released on August 8, 2025, demonstrates strong overall performance with a focus on efficiency and reliability. The model performs among the fastest models, ranking in the top tier for speed (65th percentile). It offers moderate pricing, positioning it competitively within the market (28th percentile). Jamba Large 1.7 exhibits excellent reliability with a 92% success rate, indicating consistent and usable responses. In terms of specific benchmarks, Jamba Large 1.7 excels in Email Classification (99.0% accuracy, 79th percentile) and Ethics (99.0% accuracy, 48th percentile), showcasing its ability to handle structured classification tasks and ethical reasoning effectively. Its General Knowledge (93.0% accuracy) is also solid, though it ranks in the 31st percentile. A notable weakness is its performance in Hallucinations, with only 52.0% accuracy (9th percentile), suggesting a tendency to generate information rather than acknowledge uncertainty. Mathematics (40.0% accuracy, 16th percentile) and Instruction Following (44.0% accuracy, 34th percentile) also present areas for improvement. Despite these, its hybrid SSM-Transformer architecture and 256K context window contribute to its overall efficiency and improved grounding.