GPT Image 2.6 Takes the Top Two Spots: Sunburst Leads on Quality, Flare Is Twice as Fast

On November 7, 2026, independent benchmarking group Artificial Analysis published its latest results: OpenAI's GPT Image 2.6 has taken the top two places on both the text-to-image and image-editing leaderboards. Sunburst and Flare came first and second on both boards — the first third-party verdict three days after 2.6 launched.

Compared with the previous generation, Sunburst scores 32 Elo higher than 2.5 on text-to-image and 10 Elo higher on editing; Flare is 20 and 24 Elo ahead, respectively. Artificial Analysis builds its boards from blind user votes: participants do not know which model produced an image, and their preferences are converted into Elo scores.

Speed or quality — this time you don't have to choose

Speed is the most striking change this time. Flare's median text-to-image time is about 30 seconds, nearly twice as fast as 2.5's roughly 55 seconds, with editing at about 32 seconds; Sunburst is far slower, at about 2 minutes 15 seconds per image — slower than 2.5 — but that cost buys it both first places. Artificial Analysis concludes that Flare matches the image quality of 2.5 Sunburst in about a quarter of the time, making it the only model in the most attractive quadrant of its quality-versus-speed chart.

A leaderboard win is not a complete victory. Elo scores reflect the preferences of blind-test voters, not a direct measure of text accuracy, detail fidelity, or production readiness; and this is only Artificial Analysis's board — on LMArena's text-to-image leaderboard, 2.6 Sunburst currently sits second, with voting still preliminary. For most users, the takeaway is straightforward: Flare is the better value for everyday generation and editing, while Sunburst is for when you want the best quality and are not in a hurry. ChatGPT switched all tiers to Images 2.6 on November 4. How to use the new GIF and reference-image features