InternVL2.5 78B

InternVL2.5 78B

Quick answer: InternVL2.5 78B is OpenGVLab's flagship open-weight multimodal model, part of the InternVL2.5 family built on a progressive scaling strategy, and the top-scoring open-source model on the MTVQA multilingual text-centric visual QA benchmark.

At a Glance

Where InternVL2.5 78B leads

• Strong multilingual OCR and scene-text comprehension, outperforming larger proprietary models on MTVQA
• Fully open weights, enabling self-hosted deployment and fine-tuning

Where it lags

• Requires substantial GPU infrastructure to self-host at 78B parameters
• Trails top proprietary models on some general reasoning benchmarks

Best for: teams needing strong open-weight multilingual visual document/OCR understanding with full deployment control.

What InternVL2.5 78B Is

InternVL2.5 78B is part of the InternVL2.5 series from OpenGVLab (Shanghai AI Laboratory), an open-weight multimodal LLM family built on a progressive scaling strategy that expands performance boundaries of open-source vision-language models. It combines a large vision encoder with an LLM backbone to handle high-resolution image and multi-image understanding tasks.

Specifications