Introducing ChatGPT Images / GPT Image 1.5 (OpenAI announcement)

TL;DR

OpenAI’s launch tweet for ChatGPT Images, a new flagship image generation model exposed in the API as GPT Image 1.5. The announcement is a one-liner capability list: stronger instruction following, precise editing, detail preservation, and ~4× faster generation than the prior model. Rolling out to all ChatGPT users on Dec 16, 2025 and to the API as gpt-image-1.5. No architecture, no benchmarks, no per-axis numbers — this is a product launch, not a tech report. Historically interesting as the immediate predecessor to ChatGPT Images 2.0 / gpt-image-2 (April 2026).

Key claims

New image generation model branded “ChatGPT Images” in product, “GPT Image 1.5” in the API [tweet].
Four marketed capability axes: instruction following, precise editing, detail preservation, ~4× generation speedup over the previous model [tweet].
Available to all ChatGPT tiers (not Plus-gated) and via OpenAI API from Dec 16, 2025 [tweet].
3.2M views on the announcement tweet within hours [tweet metadata].

Method

Not disclosed. The tweet ships a video example and a single still image, with no description of the underlying generator family (diffusion vs autoregressive), training data, parameter count, or guidance mechanism. The “4× faster” claim has no reference baseline named in the tweet.

Results

No benchmarks. No FID, no human-eval, no GenEval, no DPG-Bench numbers in the announcement.
“4× faster than before” is the only quantitative claim, and “before” is unspecified — presumably gpt-image-1 (the March 2025 model).

Why it’s interesting

This is the direct ancestor of Introducing ChatGPT Images 2.0 (the April 2026 release of gpt-image-2, with native “Thinking mode”). The pair pin down OpenAI’s pre-2.0 baseline: 1.5 is positioned as a pure capability+latency upgrade with no reasoning loop, whereas 2.0 four months later folds in [[thinking-with-modalities]] (web search, layout planning, self-verification). For competitive context, this lands in the same closed-API frontier-image cohort as Nano Banana 2 (Nano Banana 2 / Gemini 3.1 Flash Image (Google product announcement)) — both are tweet-only product launches with no technical detail, in sharp contrast to the open releases of the same window like FLUX.2 [klein]: Towards Interactive Visual Intelligence, Qwen-Image-2.0, and HunyuanImage 3.0 Technical Report that ship full architecture and ablations. The 4× speedup claim is the one quantitative number worth tracking — it suggests OpenAI invested in inference optimization between gpt-image-1 and gpt-image-1.5, which would be consistent with a diffusion-distillation or few-step-generation backbone, but the tweet doesn’t say.