Grok 2 Vision alternatives
Models that cover what Grok 2 Vision does, closest first. All of them take the same request, so trying one is a one-word change.
General vision model with strong handwriting and layout understanding at a low price.
General vision model. Good at messy scans and following extraction instructions.
General vision model with careful reading of long, dense documents.
Mistral's open-weight vision models, Pixtral 12B and Pixtral Large. Mistral's API no longer serves them, so they run self-hosted.
Shanghai AI Lab's open vision-language family (1B to 241B, August 2025), strong on OCR and document benchmarks.
8B vision model from OpenBMB (August 2025) built to run on devices, with strong OCR and document parsing.
Dashed tags are things Grok 2 Vision does not support. Model IDs: google/gemini-flash, openai/gpt, anthropic/claude, mistral/pixtral, opengvlab/internvl, openbmb/minicpm-v.