News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM

Images in

Reads photos, charts and documents. 118 models on EveryLLM so far.

Models with image input can read photographs, screenshots, charts, diagrams, handwriting and scanned documents, and answer questions about them. An image encoder, often a vision transformer, cuts the picture into small patches and turns each into a vector that the language model treats like a token. Vision is now standard in frontier models and common in open ones.

How vision was added

Early vision-language models joined a pre-trained image encoder to a pre-trained language model. DeepMind's Flamingo (2022) did this with special cross-attention layers, and LLaVA (2023) used a simple projection trained on instruction data written by GPT-4. GPT-4 accepted images from March 2023, a feature that reached ChatGPT users in September 2023. Claude 3 (March 2024), Llama 3.2 (September 2024) and Gemma 3 (March 2025) brought vision to more users.

Uses and limits

Document understanding is the largest practical use: reading invoices, forms, tables and slides, and turning scanned pages into text. Models also describe images for blind and low-vision people, starting with a partnership between OpenAI and the Be My Eyes app in 2023. Weak spots include counting objects, reading small print, precise spatial reasoning and spotting edited photographs, and models can be misled by instructions hidden in images.

Famous images in models

Still to be written: GPT-4o.

Further reading

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.