OpenAI releases ChatGPT Images 2.5 with faster, more accurate image generation and new Sketch-based input for in-chat drawing
Feature update ● Confirmed 95% confidence first seen
OpenAI introduced ChatGPT Images 2.5, an update to its image-generation models used in ChatGPT and related APIs. The release adds faster generation and editing, along with a Sketch input that lets users draw a doodle in chat for image generation, plus model variants and updated tooling for reference images.
Decision brief
- What changed
- OpenAI released ChatGPT Images 2.5 for ChatGPT and the GPT-Image API, updating its image-generation models and adding sketch-based image input so users can generate images from in-chat doodles alongside optional text instructions. The API update also introduced new model IDs, including gpt-image-2.5-sunburst and gpt-image-2.5-flare, with related tooling support for reference images.
- Why it matters
- This expands image generation from text prompting toward a more directed workflow using sketches and reference photos, which can matter for teams that need faster creative iteration or tighter control over visual outputs. For product and technology leaders, the API and tooling changes create a concrete integration decision: whether to adopt the new models to improve speed, multi-turn instruction-following, and subject preservation in image-based features. For operating leaders, the release may reduce friction in design and content workflows, but only if the new behavior is reliable enough for production use.
- Evidence
- The core announcement comes from OpenAI’s product blog, which states that ChatGPT Images 2.5 improves image generation and supports turning sketches and reference photos into polished images. The Verge independently confirms the new in-chat Sketch workflow, while Simon Willison’s write-up adds the reported API model IDs and claims around faster responses, better multi-turn instruction-following, reference-photo preservation, and CLI support; these points are consistent across the coverage where they overlap.
- What remains uncertain
- The coverage does not establish benchmarked performance, pricing, availability by tier or region, or how materially better outputs are in real production settings versus earlier models. It is also unclear how well sketch input and reference-photo handling generalize across enterprise use cases, or what governance, copyright, and safety controls apply in practice.
- Monitor next
- Watch for OpenAI to publish pricing, rollout details, or independent side-by-side evaluations of ChatGPT Images 2.5 and its new API models in production workflows.
Analytical support, not advice — assumptions and open questions stated above.