Skip to main content
Rase AI Intelligence Newsletter

Models & assistants Model or AI service

InternVL3 provides downloadable vision-language models

OpenGVLab released InternVL3 as a family of downloadable vision-language models. The models are designed to answer questions about visual inputs as well as text. [1][2]

Model access
The project lists several downloadable InternVL3 checkpoints, including the cited 78B variant. [1][2]
License
The 78B card states MIT for the project and separate Qwen License terms for its pretrained Qwen component. [2]

What the model does

The researchers introduced InternVL3 on April 11, 2025 and linked multiple sizes, including a 78B checkpoint. The project describes training that pairs visual and text inputs from early stages. [1][2]

Why it matters

Teams can evaluate a model on how well it reads the contents of a visual input and how well it reasons about those contents. A correct-looking response can still misread a label, axis or number. [1][2]

Who can use it

Research and application teams can compare InternVL3 checkpoints for tasks using documents, images or video. Model choice affects the hardware and input processing they need. [1][2]

Key dates

  1. InternVL3 family introduced [1]

Ways to use it

For a document image, ask the model to extract a value and identify where it found it. Check both the value and the source region against the original image. [1][2]

Model limitations

Licenses and serving needs depend on the checkpoint. The 78B card states an MIT project license and identifies a pretrained Qwen component under the Qwen License. [2]

Sources and verification

Catalog evidence checked How we verify records

  1. Primary source OpenGVLab’s dated InternVL3 family introduction and checkpoint list
  2. Primary source OpenGVLab/InternVL3-78B

Related profiles

Models & assistants

GPT-6.2 Astra and Sol

OpenAI

Astra targets demanding projects. Sol targets everyday tasks. Both support text and image input in the API.

High confidenceEvidence checked

Models & assistants

MAI Code 2 and Code 2 Flash

Microsoft AI

Code 2 targets complex software work. Flash targets frequent edits, explanations and routine fixes.

High confidenceEvidence checked

Models & assistants

MAI Realtime 1

Microsoft AI

Users can try a MAI Labs demo. Developers can register for a limited Foundry preview.

High confidenceEvidence checked