AI context window.
What is an AI context window?
The context window of a large language model is the maximum number of tokens the model can take as input in a single inference call, including system prompt, conversation history, retrieved documents, tool definitions, and the user's current message. It also bounds the output side — the model cannot produce more tokens than fit in the remaining space. In 2026, frontier model context windows are: Claude Opus 4.5 (1 million tokens), GPT-5 (400K tokens with 1M for select tasks), Gemini 2.5 Pro (2 million tokens), Llama 3.3 70B (128K). Effective context — where the model actually retrieves accurately — is usually shorter than nominal context, measured by needle-in-a-haystack benchmarks. Larger contexts enable long-document Q&A (200-page PDFs, whole codebases), longer agent sessions, and richer RAG, but cost more per call and increase latency. Scroll down for the per-model context-window comparison.
The context window is the maximum number of tokens an AI model can read in a single request — combined input + output. Larger context = the model can consider more text at once.
As of 2026: Claude Opus 4.5 = 200K, GPT-5 = 256K, Gemini 2.5 Pro = 2M (largest). 1 token ≈ 0.75 English words, so a 200K window holds roughly 150K words (≈ 600 pages of book text).
Jarvis (getjarvis.eu) sends very large documents (>100K tokens) to whichever frontier model has the widest context window, and complex code review to the strongest reasoning model at default sizes.
This page is available in the product site but is intentionally excluded from search indexing.
Glossary
