GPT Image 2.6 Flare

• - 🏢 OpenAI image generation and editing model, released November 2026.
• - 🆕 Successor to GPT Image 2.5 Flare in the Flare line.
• - 👁️ Accepts both text and image inputs to produce images.
• - 🔧 Supports a complexity setting and markup-guided edits.
• - 🆕 Can output SVG and animated GIF formats.
• - 💬 Documented for interactive generation and editing workflows.
• - 🎯 Paired with a Sunburst tier in the same 2.6 generation.
• - 🔒 Closed-weights, API-only via OpenAI's images endpoints.

OpenAI

OpenAI is an American artificial intelligence research organization headquartered in San Francisco, structured as both a for-profit public benefit corporation and a nonprofit foundation. The lab developed the GPT family of large language models, the DALL-E…

GPT Image 2.6 Flare is OpenAI's November 2026 image model, part of the GPT Image 2.6 generation that also includes GPT Image 2.6 Sunburst. According to OpenAI's API documentation, it generates and edits images from text and image inputs, and is intended for interactive image generation and editing workflows. Dedicated editing entries, GPT Image 2.6 Flare Edit and GPT Image 2.6 Sunburst Edit, expose the same family for inpainting-style tasks.

Relative to its direct predecessor GPT Image 2.5 Flare — and the earlier GPT Image 2 and GPT Image 1.5 — the clearest documented changes are in controllability and output formats. OpenAI's docs for this model highlight a complexity setting, markup-guided edits, SVG output and animated GIF output, controls that were not part of the documented surface for earlier GPT Image releases. Markup-guided editing in particular lets a user annotate a reference image rather than describing every change in prose.

Within the 2.6 generation, Flare and Sunburst are offered as two distinct model identifiers through OpenAI's image generation and edit endpoints, so applications can pick a tier per workload rather than switching families. Because the weights are closed and OpenAI has not published head-to-head quality figures for this release, no benchmark numbers are quoted here; capability claims above are limited to the provider's own documentation.