AI Model Context Window Cheat Sheet

Multimodal flagships; fast tokenizer.

Cheaper, same context window as 4o.

Reasoning models; large context for code/reasoning.

1M-token context for long-document analysis.

Legacy reasoning series; smaller output cap.

Highest reasoning quality; 200K context.

Balanced speed and capability.

Fast and cheap; same context as larger siblings.

Widely deployed; strong coding agent use.

Legacy Claude 3 family.

1M context; strong long-context retrieval.

Fast multimodal with 1M context.

Up to 2M tokens for video/audio docs.

First mainstream 1M-context model.

Multimodal MoE; long context for local.

Strong coding and reasoning.

Efficient instruction model.

128K standard across all sizes.

Tiny edge models with long context.

128K context across dense and MoE variants.

Strong open models; 128K context.

Vision-language with long context.

Commercial flagship; 128K window.

Open-weight small models with 128K.

Vision model with 128K context.

Code-specialized; 32K context.

MoE generalist; 64K context.

Open reasoning model; 64K context.

Earlier version with 128K claim (often 64K practical).

128K context for chat and coding.

Prior generation with same 128K window.

Context limited by VRAM, not model; reduce KV cache with quantization.

Serves full-context models if GPU memory allows.

Offload-aware; effective context depends on RAM/VRAM.

Unified memory; large context possible on Macs with 32-128 GB RAM.