ChatGPT capabilities

Real-time, low-latency speech-to-speech conversational mode using the GPT-4o-class realtime audio model. Supports interruptions, tone/emotion variation, accents, and live language interpretation.

Side-by-side collaborative editing interface for iterating on writing and code within ChatGPT. NOTE: per GPT-5.5 release notes, the dedicated Canvas side-panel was removed from GPT-5.5 Instant and GPT-5.5 Thinking on May 28, 2026; its writing/code editing folded into in-chat writing blocks and code blocks. Paid users retain the legacy Canvas only on older not-yet-sunset models — GPT-4.5 was removed June 26, 2026 and GPT-5.2 on June 12, 2026 (conversations on removed models auto-continue on GPT-5.5, without the Canvas panel), leaving o3 (retires August 26, 2026) as the main remaining legacy holdout.

Agentic mode that gives ChatGPT a virtual computer (visual browser, text browser, terminal, direct API access) to autonomously complete multi-step web and document tasks on the user's behalf.

ChatGPT capability that writes and runs Python in a sandboxed environment to do real computation: data cleaning and statistical analysis, math, file processing, and generating static and interactive charts and tables from uploaded data. Vendor docs now surface it as 'Data analysis' (formerly Advanced Data Analysis / Code Interpreter); it is frequently conflated with plain file upload, but the upload only puts a file in context, whereas this capability actually executes code against it and returns generated artifacts the user can download.

Cloud-based software engineering agent embedded in ChatGPT (and as a separate Codex app/CLI/IDE extension). Spins up sandboxed cloud environments to write features, fix bugs, answer codebase questions, and open pull requests in parallel.

Integrations ('Apps', renamed from connectors on December 17, 2025) that connect ChatGPT to external services like Google Drive, OneDrive, Slack, and other apps — spanning interactive in-chat apps, search/reference connectors, deep-research sources, synced knowledge, and write actions.

User- or workspace-installed connections that let ChatGPT call approved external tools and retrieve data from third-party services. Built on the Model Context Protocol (MCP); unified under the 'Apps' branding since Dec 17, 2025 (apps + connectors).

User-created or community-built custom AI assistants with specific instructions, knowledge files, and optional API actions.

Persistent user-level instructions that ChatGPT applies to its responses across all chats.

Multi-step agentic research mode that plans, searches the web (and optionally connected apps/MCP sources), and synthesizes a long-form cited report. Designed for tasks taking 5–30 minutes of autonomous browsing.

Upload files to conversations for analysis, summarization, data extraction, and code execution.

Multi-person conversations (up to 20 humans) sharing a single ChatGPT thread. ChatGPT decides when to chime in versus stay quiet and can be invoked explicitly with @ChatGPT.

Native image generation and editing inside ChatGPT, powered by ChatGPT Images 2.0 (gpt-image-2). Replaces DALL-E as the default pipeline (legacy DALL-E is still reachable via the DALL-E GPT) and the discontinued Sora image flow; covers text-to-image, image-to-image edits, and selection-based inpainting.

Persistent memory that ChatGPT builds across conversations to retain user context and preferences.

Consumer and team subscription tiers for ChatGPT as of July 2026: Free, Go, Plus, Pro ($100), Pro ($200), Business, and Enterprise. Tiers differ in model routing, message/Deep-Research/Agent quotas, context window, and admin/compliance features.

Named workspaces that group conversations, files, and custom instructions for a long-running effort.

Public, unlisted URLs that publish a snapshot of a conversation (or a single response) so anyone with the link can view it and optionally import it into their own chat history.

Text-to-video and image-to-video generation with synchronized audio, powered by Sora 2 (and the higher-quality Sora 2 Pro). NOTE: OpenAI removed video generation from ChatGPT on April 26, 2026 and shut down the standalone Sora app the same day, so this capability is no longer available inside ChatGPT; only the deprecated Sora 2 Videos API remains, scheduled to shut down September 24, 2026 with no replacement.

Scheduled and recurring jobs that ChatGPT runs for the user at a specified time, delivering results via push notification or email even when the user is offline.

Real-time web search integrated into ChatGPT conversations for retrieving current information.