Introducing Images 2.6

openai.com2 min read

OpenAI has introduced Images 2.6 across ChatGPT, Codex, and its API, adding new motion capabilities through GIFs, 50 percent faster generation speeds, and greater edit consistency. The update introduces GPT-Image-2.6 Flare and Sunburst models in the API for developers, alongside features such as image markup, SVG exports, and a new complexity parameter. Open article →

Notable comments

Reception is mixed: some users see GIF/animation and SVG-oriented improvements as belated but promising, while others report worse quality/consistency than prior versions, design-control gaps, and potential pricing/token regressions for typical workflows.

HN · bsboxe Provides concrete pricing/token usage comparisons showing 2.6 can use ~4x more tokens than 2.5 at low/medium quality (e.g., 1024x1024: 196→823 tokens low), undermining the “faster generation” narrative for many settings.

HN · minimaxir Points out animation is likely built on the common 'grid trick' of generating individual frames and notes frame-to-frame adherence/noise remains a fundamental issue; also raises skepticism about SVG support given how image models work.

HN · jjcm First-hand A/B testing claims 2.6-flare rolls back to 2.5-flare because 2.6 is either 'too simplistic' or yields outputs that break in other cases, i.e., perceived quality regressions despite the release.

More on this topic