InternLM-XComposer-2.5 is a versatile large vision language model that supports long-contextual input and output. It excels in text-image comprehension and composition tasks, achieving capabilities comparable to GPT-4V with a 7B LLM backend. Trained on 24K interleaved image-text contexts, it can extend to 96K long contexts via RoPE extrapolation. The model features three major upgrades in vision-language comprehension: ultra-high resolution understanding, fine-grained video understanding, and multi-turn multi-image dialogue. It also supports two applications using extra LoRA parameters: crafting webpages and composing high-quality text-image articles. IXC-2.5 has been evaluated on 28 benchmarks, outperforming existing open-source models on 16 benchmarks and surpassing or competing closely with GPT-4V and Gemini Pro on 16 key tasks. The model is publicly available at https://github.com/InternLM/InternLM-XComposer. It supports audio input and output using open-source tools and has been tested on various video and image benchmarks, demonstrating strong performance in video understanding, high-resolution image analysis, and webpage generation. The model is capable of generating high-quality text-image articles through a scalable pipeline that integrates supervised fine-tuning, reward modeling, preference data collection, and DPO alignment. IXC-2.5 is designed to handle diverse multimodal tasks and has been evaluated on various benchmarks, showing competitive performance with closed-source APIs and open-source models. The model is capable of generating high-quality text-image articles and has been tested on various benchmarks, showing competitive performance with closed-source APIs and open-source models.InternLM-XComposer-2.5 is a versatile large vision language model that supports long-contextual input and output. It excels in text-image comprehension and composition tasks, achieving capabilities comparable to GPT-4V with a 7B LLM backend. Trained on 24K interleaved image-text contexts, it can extend to 96K long contexts via RoPE extrapolation. The model features three major upgrades in vision-language comprehension: ultra-high resolution understanding, fine-grained video understanding, and multi-turn multi-image dialogue. It also supports two applications using extra LoRA parameters: crafting webpages and composing high-quality text-image articles. IXC-2.5 has been evaluated on 28 benchmarks, outperforming existing open-source models on 16 benchmarks and surpassing or competing closely with GPT-4V and Gemini Pro on 16 key tasks. The model is publicly available at https://github.com/InternLM/InternLM-XComposer. It supports audio input and output using open-source tools and has been tested on various video and image benchmarks, demonstrating strong performance in video understanding, high-resolution image analysis, and webpage generation. The model is capable of generating high-quality text-image articles through a scalable pipeline that integrates supervised fine-tuning, reward modeling, preference data collection, and DPO alignment. IXC-2.5 is designed to handle diverse multimodal tasks and has been evaluated on various benchmarks, showing competitive performance with closed-source APIs and open-source models. The model is capable of generating high-quality text-image articles and has been tested on various benchmarks, showing competitive performance with closed-source APIs and open-source models.