InferX Beta — Serverless GPU Inference Platform, Built for Agent-Native Workloads

allenai/Molmo2-8B is an 8B multimodal vision-language model from the Allen Institute for AI, built on Qwen3-8B with a SigLIP-2 vision backbone, designed for image, multi-image, and video understanding tasks such as captioning, counting, tracking, and visual grounding, delivering state-of-the-art performance among open-weight models while remaining efficient for production deployments.