Ultra-long context, Qwen2-VL outperforms GPT-4o, new open weights Text to Video model, Eagle multimodal large language model, fastest AI inference and more