Six image models become one: OpenAI's deprecation list dates the end of every generator it ever sold

OpenAI's deprecation page is not an announcement, which is why the scale of what it describes is easy to miss. Read end to end, it schedules the retirement of every image generator the company has sold, in three notices, leaving a single model standing. The first notice, dated 14 November 2025, covered dall-e-2 and dall-e-3; both were removed from the API on 12 May 2026. The official DALL·E GPT inside ChatGPT outlived them and goes dark on 30 August 2026. The second, dated 22 April 2026 — one day after GPT Image 2 was released — deprecated gpt-image-1, with a shutdown on 23 October 2026. The third, dated 2 June 2026, covers gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest, all switched off on 1 December 2026. Every notice names the same replacement: gpt-image-2. For developers the migration is mostly favourable. GPT Image 2 drops the three fixed shapes of its predecessor and accepts any resolution up to a 3840-pixel edge, provided both sides are multiples of 16, the ratio stays inside 3:1 and the total lands between roughly 0.65 and 8.3 megapixels. Reference images are always processed at full fidelity, and generation can be streamed with up to three partial previews. One thing does not survive the move. GPT Image 2 does not support transparent backgrounds — a request with background set to transparent is rejected — while gpt-image-1 did. Anyone generating logos, icons or overlay artwork loses a feature on 23 October and has no in-house replacement for it. The billing model changes too. Where gpt-image-1 charged per picture, from USD 0.011 for a low-quality square to USD 0.25 for a large high-quality one, gpt-image-2 bills tokens: USD 8 per million on input, USD 2 cached, USD 30 per million on output, halved in the Batch API. OpenAI's own example puts a high-quality 1024×1024 image at about USD 0.21 — more than the equivalent square on the old model, and a difference that only shows up once volume is counted.