GPT‑Image‑2 tops leaderboards

OpenAI’s GPT‑Image‑2 swept Arena’s image leaderboards after the model’s April 21, 2026 launch, topping both text‑to‑image and editing tracks. (openai.com) Arena’s Text‑to‑Image leaderboard shows GPT‑Image‑2 at 1,512 Elo, a +242‑point lead over the nearest rival, Nano Banana 2 (Google’s Gemini 3.1 image model). (arena.ai) OpenAI’s model also posted 1,513 Elo on single‑image editing and 1,464 Elo on multi‑image editing, putting it #1 across every Image Arena category within hours of release. (bestphoto.ai) Arena builds its rankings from blind, pairwise human‑preference votes aggregated into an Elo‑style rating, not from synthetic pixel‑level metrics. (arena.ai) By Elo convention a 100‑point gap implies roughly a 64% win rate for the higher‑rated entry, so a 242‑point margin represents a substantially higher head‑to‑head preference on Arena’s scale. (lambdafin.com) OpenAI describes ChatGPT Images 2.0 (the gpt‑image‑2 model) as adding a pre‑generation “thinking” planning phase, near‑perfect multilingual text rendering, multi‑image consistency, and availability in ChatGPT, the API and Codex. (openai.com) Arena’s nearest competitor is Google’s Nano Banana 2 (Gemini 3.1 Flash Image), and independent analysts caution that crowdsourced leaderboards can reflect prompt selection, voter demographics, and statistical noise. (officechai.com, staituned.com) Arena traces its roots to the Chatbot Arena project (LMSYS) launched in 2023 and rebranded as Arena in 2026, making these leaderboards a visible, community‑driven benchmark watched by vendors and developers. (lmsys.org, arena.ai) “Images are a language, not decoration,” OpenAI wrote in its product notes describing Images 2.0 — a framing the company has paired with the model’s rapid top‑of‑board results. (petapixel.com)