GPT Image 2.5 launched September 8, 2.6 on November 4 — only 8 weeks apart. Both generations ship as Sunburst + Flare pairs at identical token rates. The reasons to upgrade are three new features and higher scores; the reasons not to are speed (Sunburst got slower) and workflows (Responses/Batch aren't supported).
The official changelog is clear: 2.6 adds animated GIF generation (2–16 frames, default 9 × 200ms = a 1.8s loop, frame-by-frame editable), SVG export (API-only, two-stage), and the complexity planning parameter (controls how much planning happens before rendering, default high, independent of quality, no separate fee). The easy confusion is markup editing — it went to supported 2.5 models at the same time, so it's not a 2.6 exclusive; if markup is your only reason to upgrade, you don't need to switch generations.
Artificial Analysis (read Nov 13–14): T2I 2.6 Sunburst (max) 1261 (#1) vs 2.5 Sunburst (max) 1229 (#2); editing 1210 (#1) vs 1201 (#3) — 2.6 leads on both boards. LMArena (Nov 12 board): T2I 2.5 Sunburst 1429 (#1, 63,818 votes) vs 2.6 Sunburst (Medium) 1406 (#2, 2,697 votes); editing 1519 (#1) vs 1476 (#3) — 2.5 still leads, but 2.6 has only a few thousand votes, is marked Preliminary, and has wide confidence intervals (±10–13), so it can't be directly compared against 2.5's six-figure vote counts. Rapidata's overall board (Nov 13 sync): Sunburst 2.6 #1 (1308), Flare 2.6 #2 (1284), 2.5 Sunburst #3 (1212).
QuantSlant's 12 identical prompts (medium): 2.5 Sunburst median 17.9s, 2.6 Sunburst 26.3s, 2.6 Flare 8.0s — "50% faster" holds only for Flare, and Sunburst 2.6 was slower than 2.5 Sunburst on all 12 prompts. OverChat agrees: Flare travel ad at low 9.7s vs 2.5 Flare's 15.0s (1.5x), living-room edit 16.6s vs 21.4s (1.3x); 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. The official announcement says "50% faster" without naming the comparison baseline.
Token rates are identical (text $5, image input $8, image output $30 per 1M tokens), but per-image token use at low/medium quality is 4.2x/3.3x that of 2.5: low $0.02469 (2.5: $0.00588), medium $0.04347 (2.5: $0.01317); high and above cost the same. QuantSlant's same-prompt comparison (medium): 2.5 Sunburst 343 tokens ($0.011) vs 2.6 Sunburst 1,132 tokens ($0.034), about 3x the bill; a three-round editing flow cost $0.046 / ~27s on 2.6 vs $0.022 / ~19s on 2.5. Another price difference: 2.5 has Batch pricing (image output at half price, $15/1M); 2.6 has none. See our pricing guide for details.
2.6 supports only the Image API's generations/edits endpoints; Responses API image tools, Chat Completions and Batch API are all marked unsupported. If your current integration depends on the Responses API's image tools or Batch processing, upgrading to 2.6 means rewriting the integration — not just swapping a model ID. Whether 2.6's cached rates are realizable is also unverified (the guide's caching section covers only 2 and 2.5 inside the Responses API).
Need GIF, SVG or complexity: upgrade — 2.5 doesn't have them. Need the fastest generation: upgrade to Flare — as fast as ~6s at low quality (best case), about 1.5x faster than 2.5 Flare. Using Sunburst for multi-round editing, depending on Responses/Batch workflows, or sensitive to low-quality per-image cost: wait — Sunburst 2.6 is slower, low-quality images cost 4.2x/3.3x more, and there's no Batch price; decide after the SVG reliability follow-ups and more speed data. ChatGPT users need do nothing: ChatGPT, Codex and the API all got 2.6 on Nov 4.
Reference: QuantSlant — GPT Image 2.6 vs 2.5 (rewritten and organized by ModelBrief; facts per the source)