Zyphra Releases Zamba2-VL: Hybrid Mamba2–Transformer Vision-Language Models

Zyphra has launched Zamba2-VL, a family of open vision-language models (VLMs) in 1.2B, 2.7B, and 7B parameters. These models utilize a hybrid State Space Model (SSM)–Transformer backbone, replacing dense Transformers to achieve competitive accuracy with significantly lower latency. Zamba2-VL shows particular strengths in visual counting and document understanding, offering up to an order of magnitude faster Time-to-First-Token (TTFT) compared to Transformer baselines.