Meituan LongCat's WorldCast benchmark: GPT-Image-2.5 tops at 72.8, open-source lags
meituan-longcat · hf · 2026-10-16
Meituan's LongCat team introduces WorldCast, a benchmark for multi-step world state prediction in visual generation, along with a four-level capability hierarchy (L1 pixel-level control to L4 real-world context transformation).
- Scale: 320 cases, 1352 states, 12 models, scored on visual consistency, causal plausibility, and task completion.
- Results: GPT-Image-2.5 Sunburst leads at 72.8 overall, followed by Seedance-2.0 (47.2) and Nano Banana Pro / Gemini-3-Pro-Image (42.8); open-source models like Emu3.5 (12.3) trail badly on reasoning-heavy visual forecasting.
- Task style: e.g. generating a four-step toy assembly process from parts photo plus manual, preserving the work surface, component appearance, and temporal continuity.
More from Multimodal
- 1986 to 2026 in one session: outpaint, modernize and 4x upscale, session file included — juanpablogc · 2026-10-16
- Three resources to get up to speed with FLUX 3 — ariG23498 · 2026-10-16
- Watch: Claude Opus 5.5 builds a Tesla Roadster from a single prompt — TAbrodi · 2026-10-16
- Midjourney v8.2 turns out striking indigo ink-wash art with sref 1848359068 — azed_ai · 2026-10-16
- "AI television from another dimension": robot grooms a chihuahua on air — BotLordBotflix · 2026-10-16
- 108M-parameter text-to-image denoiser Surjo-Image-Preview lands on Hugging Face — Consistent-Rule-3720 · 2026-10-16