GPT-5¶

OpenAI's 2026 flagship multimodal model. Successor to GPT-4o. Shipped with a 128K context window, text + vision + audio input modalities, and a pricing structure that sits slightly above Claude Sonnet 4 on input but comparable on output.

Notable features¶

• Multimodal from day one — text, vision, and audio in a single endpoint.
• 16K output cap — double the 8K output ceiling of most contemporaries; useful for long-form structured generation.
• ARC-AGI 2 score — 16.4%, the first major model to score meaningfully on the upgraded ARC-AGI benchmark (most models are in the low single digits).

Connections¶

• [[OpenAI]] — the provider
• [[MultimodalModels]] — primary category
• [[ARC-AGI 2]] — novel benchmark