GPT Image 2.6 comes in two variants: Flare and Sunburst. Officially, Flare is "fast iteration" and Sunburst is "detailed visual work". The most important fact first: token prices are identical for both, and token use at the same settings measures the same — three sources measured identical bills. The choice trades time for detail, not money.
QuantSlant's Nov 5 test (3:2 bakery prompt, via OpenRouter): low 7.4/19.0s, medium 7.4/23.7s, high 10.1/39.2s, max 24.7/118.3s (Flare/Sunburst). Median across 12 identical prompts (medium): 2.6 Flare 8.0s, 2.6 Sunburst 26.3s. OverChat agrees: Flare travel ad at low 9.7s, living-room edit 16.6s, Sunburst 2.6 at 43–61s per image. Treza measured Flare at ~6s low / ~40s max, Sunburst at ~17s low / ~166s max. OpenAI says Flare is as fast as ~6s at low quality — note that's a best case, not an average.
Officially, Sunburst delivers higher quality at medium/high and stronger multi-round editing consistency. Independent scores support that positioning: on AA's boards Sunburst (max) scores T2I 1261 (#1) and editing 1210 (#1), Flare (max) T2I 1226 (#3) and editing 1206 (#2) — the gap is small, but Sunburst leads on both boards. On Rapidata's overall board: Sunburst 1308 (#1), Flare 1284 (#2); on the text-coherence dimension Flare's 1348 (#1) actually beats Sunburst's 1254 (#2).
The official announcement says Flare has improved layout quality, suited to posters, product shots and other fast-turnaround work. Treza's test adds a detail: Flare at low quality performs best with short, simple prompts; on complex prompts low-quality Flare tends to drop details. Sunburst's advantage is multi-round editing — OpenAI says its consistency across rounds is stronger, holding subject and composition better through an editing flow.
Fast turnaround, batch exploration, layout work: pick Flare — seconds per image at low/medium quality, the same bill as Sunburst. Final artwork, multi-round editing, detail-dense scenes: pick Sunburst at medium/high, accepting tens of seconds to ~2 minutes per image. Note there's no Flare/Sunburst selector in ChatGPT, and OpenAI hasn't disclosed which variant ChatGPT calls; API users must specify the model ID in the request (gpt-image-2.6-flare / gpt-image-2.6-sunburst).
QuantSlant's 12-prompt medians: 2.5 Sunburst 17.9s, 2.6 Sunburst 26.3s — Sunburst 2.6 was slower than 2.5 Sunburst on all 12 prompts. OverChat found the same: Sunburst 2.6 at 43–61s per image vs 2.5 Sunburst's 32–42s. The authors note launch-day server congestion may have affected the data, but two independent sources agree. If you're upgrading from 2.5 Sunburst, expect better quality and new features — not faster generation. See our 2.6 vs 2.5 upgrade guide for the full analysis.
Reference: QuantSlant — GPT Image 2.6 Flare vs Sunburst (rewritten and organized by ModelBrief; facts per the source)