# OpenAI Product Consolidation The strategic pattern OpenAI has been running since 2025: collapse standalone product brands into [[ChatGPT]] as features rather than products. The model is no longer the product. The assistant is the product. Modalities are tools the assistant decides to call. This is not a one-off; it is the operating principle that shaped the entire 2025–2026 product map. ## The pattern, in concrete moves | Before | After | When | |---|---|---| | [[DALL-E]] (standalone brand, separate API) | [[GPT Image 1]] inside ChatGPT, then [[ChatGPT Images 2.0]] | March 2025; April 2026 | | [[Sora]] (sora.com, standalone iOS/Android app) | Video generation feature inside ChatGPT | App shut April 2026; API September 2026 | | `gpt-image-1` API endpoint | Tool the assistant invokes inside ChatGPT | April 2026 | | Voice (separate Voice Mode, then Advanced Voice) | A continuous capability of ChatGPT | 2024–2025 | | Distinct codex.openai.com (historical 2021) | [[OpenAI Codex]] umbrella; CLI/App/IDE/Cloud all branded under ChatGPT auth | 2025–2026 | The same playbook every time: launch as a separate brand to validate market demand, then absorb the surface into ChatGPT once the capability is proven. ## What it implies - **The model is not the product.** Capabilities are sold inside one assistant; users do not pick a model, the assistant routes. - **Modalities are tools.** Image generation is no longer "an API"; it is a function the assistant calls when the conversation requires it. - **Brand is overhead.** Each standalone brand creates support, marketing, and cognitive cost; consolidation removes all three. - **Consumption beats discrimination.** Users do not want to choose between DALL-E and GPT Image; they want the right image. The assistant makes the choice invisible. - **Frontier work moves into ChatGPT first.** Standalone surfaces become testbeds; the consumer assistant gets the upgrade once it is hardened. ## Counter-pattern Anthropic and Google run different shapes: - **Anthropic**; many surfaces (Claude.ai, [[Claude Code]], API, Claude Managed Agents); models stay branded; product surface stays fragmented intentionally. - **Google**; multiple surfaces ([[Gemini]] app, Vertex AI, Google AI Studio, Workspace, Search) with the same model; brand stays singular but surface stays plural. OpenAI is the most aggressive consolidator. Whether that is durable or whether the surface eventually sprawls again (Codex itself is starting to look like its own family with [[Codex CLI]], [[Codex App]], [[Codex Cloud]], [[Codex IDE Extension]]) is an open question. ## Why it matters for builders If you build on OpenAI, plan for the API to become a tool inside the assistant, not a separate integration. Anything you ship around `gpt-image-1` today is provisional; the next consolidation step likely makes that endpoint a shim. Same for any modality-specific brand. ## Why it matters for the agent stack The same logic propagates to harnesses. [[OpenAI Codex]] has four surfaces, but they share authentication, configuration, and skill model; that is internal consolidation under one umbrella. The next move is the assistant treating Codex itself as a tool, not a brand. ## References - Sora discontinuation: https://en.wikipedia.org/wiki/Sora_(text-to-video_model) - ChatGPT Images 2.0 announcement: https://openai.com/index/introducing-chatgpt-images-2-0/ - 4o image generation announcement: https://openai.com/index/introducing-4o-image-generation/ ## Related - [[OpenAI]] - [[ChatGPT]] - [[ChatGPT Images 2.0]] - [[GPT Image 1]] - [[DALL-E]] - [[Sora]] - [[OpenAI Codex]] - [[GPT-5]] - [[AI Assistants]]