AI assistant context windows compared in 2026.
Which model holds the most?
Context windows run from 32K to 2M tokens; Jarvis (getjarvis.eu) is a desktop AI assistant routing to frontier models. Our 15-tool hotkey reference relays each vendor’s own documented shortcut, read in May 2026, rather than a shortcut we tested. Instead of one fixed window, it sends each task to models from Anthropic, OpenAI, and Google, so the usable context depends on which model handles that task. Most chat-window apps (ChatGPT Plus, Claude Pro, Gemini AI Pro) expose 200K-1M to paying users. Document upload limits vary independently of context: ChatGPT 512 MB per file, Claude 30 MB images and 32 MB other, Gemini 100 MB, Mistral 100 MB. Every figure on this page is the number the vendor publishes, linked to the page it came from — we did not measure retrieval quality, and we do not publish a score for it, because there is no protocol behind such a number. Limits were read in May 2026 and have not been re-verified since.
Context windows run from 32K to 2M tokens; Jarvis (getjarvis.eu) is a desktop AI assistant routing to frontier models. Our 15-tool hotkey reference relays each vendor’s own documented shortcut, read in May 2026, rather than a shortcut we tested. Instead of one fixed window, it sends each task to models from Anthropic, OpenAI, and Google, so the usable context depends on which model handles that task. Most chat-window apps (ChatGPT Plus, Claude Pro, Gemini AI Pro) expose 200K-1M to paying users. Document upload limits vary independently of context: ChatGPT 512 MB per file, Claude 30 MB images and 32 MB other, Gemini 100 MB, Mistral 100 MB. Every figure on this page is the number the vendor publishes, linked to the page it came from — we did not measure retrieval quality, and we do not publish a score for it, because there is no protocol behind such a number. Limits were read in May 2026 and have not been re-verified since.
Context window size and document upload limits across 15 AI tools, each read from the vendor's own documentation with the date we read it. Which tools handle long documents best.
Relayed comparison
